Home

transcript - University of Maryland at College Park

1. The goal here is to really say Look that one is kind of like that one but this one is more like that one And overtime believe that we will have a taxonomy of social media We will actually be able to say How many different kinds of hashtags are there And not just hashtags really query terms anything that halls out a collection of social media content how is it structured And we re going to see that there are different patterns in these maps For example here is a as of this afternoon that would be us This is SSW12 this is 52 people who s recent tweets contained the SSW12 created at 1358 UTC so subtract 5 from that so that was what what is that That s like 10 o clock this morning something like that And what this caption will tell you is what am looking at What are the edges Green edges are follows and blue edges are mentions or replies And the edge thickness is an indication of how many messages were exchanged between those two people And you ll see that we have been clustered and that some of us are in different clusters and the size of the image PJ doesn t like the sizing but we re going to have a good methodological dispute later make the size to indicate follower account because believe that the location in the graph is conveying between this But think a good argument can be made that you d actually want them to be sized by between this not by follower account If you wanted to focus on only the things that are
2. a fan or maybe I m tired of a group of people who only connect A and B we share the quality that we only have two connections we connected same two people It might as well get rid of all of us replace us with a whole bridge gt gt Would you humor me and show that combined edges please gt gt can guess if you insist gt gt just wanted more discussions on the phone calls regularly and one of the features that our super programmer Tony added was to instead of having all those edges go back and forth and make the screen so busy we combine them Now you re doing bundle gt gt You didn t want bundling gt gt Yeah gt gt You do you re doing tight bundling I m saying combine We need to combine feature combine edges gt gt Is there an edge for every time someone repeats you repeats with you gt gt Yeah yeah gt gt And so and then the in the curve that just those edges won t necessarily perfectly overlap with each other gt gt No they perfectly overlap with each other gt gt Yeah gt gt So they perfectly overlap with each other gt gt Yeah gt gt No the combine edges feature is a long standing discussion between us that it puts one big gray edge between each cluster So you can see the numbers are connected between clusters and remove a great deal of the it s in the group menu gt gt have done the wrong thing I ll have to undo that feature and then redo that Yeah I we
3. actually does inaudible gt gt That may be but will still argue that that number that time series plot doesn t tell me what think it s trying to tell me It s telling me how what s the volume but not the origin of that noise And so but think you re right that you know on the eve of the election if you look at this and it shows an enormous spread and then your candidate is not being favored you may not vote for that candidate So let s see I ve got a comment here we ll come back there in a moment Yes sir gt gt Do you have any idea whether it s more important the value that s being taught or the second derivative gt gt It s really variable isn t it It s really really variable but let me show you another example Let s see think that you can see it pretty clearly in two different images made a map of the word Obama Care and then made a map of the word ACA and what really caught my eye was the idea that I ll just do Obama Care If you said Obama Care if we look at the Obama Care map what we re going to see is that two thirds of the people who are opposed to Obama Care And so this cluster and this cluster these are pretty much negative inaudible the hashtag Obama doesn t care these people don t like Obama Care These people use the word Obama Care and another thing to observe is if you look inside the profile images you ll see that the only media outlets are in this cluster Talking Points Memo Daily Feas
4. blog post or messages from any social media source And you do content analysis on that collection of tweets You are potentially making a very serious error if you don t recognize that these tweets are coming from socially defined subgroups So I ll give you an example Right now there s a bit of politics wherein a presidential elections sees and believe there are 92 days left before the election And lots of people are getting into the business trying to prognosticate the election on the basis of social media and here at election twitter com the company called Topsy is reporting horse race results on who is ahead and behind in the Twitter battle Now could try to explain how they come up with this number don t think will succeed It s allegedly the percentage of messages in Twitter that are more negative than the messages about this candidate And if you understand that you can explain it to me later but the thing I ll note about it is that one it s insanely volatile it goes up and down a lot At one point Mr Obama was at 74 and Romney at 27 they seem to be converging 18 15 But ask you this question what does this really report So let s just say and as showed you the GOP graph and is that a question or that s an answer Inaudible Remark If that s true that s unethical gt gt Oh of course But it would be laughter gt gt I m talking to you gt gt It doesn t mean they can t do it gt gt And I ll be with
5. is a map this was made by NodexXL and it s a map of relationships in Twitter And this is one of hope to show you several of these maps because it s not enough to see one you really have to see multiples in order to see in what way are they the same or different what are the categories or the varieties of them And everybody s got WiFi now right Yesterday we didn t have WiFi If you go to the website at NodeXLGraphGallery NodeXLGraphGallery all one word dot org you will see what Edward Tufte referred to as small multiples You ll see a whole bunch of liquor like images of graphs And it s my hope that simply well well gosh let s just go look at it right Laughter Why should talk about what data will show you or what page Pause And here is that page right So if you go to NodeXLGraphGallery org you ll see pictures of you How did you get in there Well you tweeted didn t you So these are all maps of various social media networks still that can be social to you there s networks They just tend to be social media networks It s what s interesting a lot now And I ll know that you as a NodexXL user you can contribute to this in the same way that pretty much anybody can upload a picture to Flickr you can upload a NodeXL graph to the graph gallery And it s important to us to get lots of these graphs in order to illustrate am running late gt gt You re on 38 minutes gt gt Oh yeah thank you Thank you very much
6. is going to be an improvement again like Group In a Box like the gridding of isolates relative in this case not that simple but the straightforward methods to improve the quality of the graph So we ve been working at a lot of graphs this is another GOP graph We see a lot of this kind of thing We see a lot of pulverization We see a lot of divergence of URLs People are not using the same words or roles And when we go and we build lots of these images we think something is emerging Categories are emerging and we can also apply this to other domains This was applied to the 2007 United States Senate voting records and there you can see clusters emerge if you say I only want to see a link between any two senators if they vote in agreement with one another Two thirds at the time are more you get the two clear divisions in US politics the east coast and west coast gt gt No laughter gt gt No and of some interest here worth noting there is Mr McCain this is 2007 Remember there was some other guy who was in the White House can t remember exactly but I ve kind of blotted out the whole thing This is Senator Obama there s Senator McCain and who are these three people Specter Snow Collins what are those three people Why are they there Cause they re high between these why are they between Cause they actually have edges into this cluster why do they have edges into this cluster Because they are traitors to their pa
7. means things like calculate your matrix and that s great you probably know that you should calculate network matrix But what you might not know is that it s probably a bad idea to do that until you ve already calculated your clusters And so what we have done is essentially built a track and you can choose not to do some things on the track but you can t change the order of the things on the track And these are the right well you will assert these are the right order to do them And in essence we can t stop you from running with scissors but we can give you the scissors with the rounded tips And so it s going to be very very hard to hurt yourself all that badly with this tool it s designed to do what we think is the right thing Now you may dispute that that s fine of course we have open code we d loved to see fourth versions or if you want to suggest different features we very much welcome your contributions on the message board and should mention to the software developers we need help Lots of work items we need work items done so let me know if you want to pick up or work item So this is the order we add up the number of edges that resemble each other so how many times did a reply we recount that Then we group by cluster we do this thing where we put them into boxes and you can see here that we have used this feature that is think an innovation and also unique to NodeXL and that is this mash up between the information visualiza
8. of value Here Nash phonetic is pointing to an article from believe it s 1933 it s part of the really the first public awareness of network analysis I m sorry gt gt Great It showed likes and dislikes at the end of the result gt gt That s right gt gt Will expect that gt gt Yeah well laughter they call them favorites but yeah yes Laughter So Jacob Marino comes from Romania in the 30s and he goes to NYU He s in Manhattan And before he has a psychotic break with reality he founds a whole new discipline of social science which is essentially the social science application of topology to human relationships and he uses the name psychological geography And he actually starts mapping out Something that just saw on the web and maybe Bernie knows what it is it s the movie network database not the gt gt Movie gallery gt gt Movie gallery so this is which characters and which movie actually will share the scene together And so you have a tie when you re on the screen together We have an example in the book a similar example of which characters interact with which characters in Les Miserables the same kind of idea Psychological geography then gets re titled Sociometry Sociometry then pretty much dies largely because Marino begins to tell people that he is in personal communication with the Almighty And this was really bad for having graduate students cause as you may know most graduates think tha
9. one domain wonder is Nancy are you here Nancy read her messages about soap operas So many so that she really understood the nature of that community But ask you what if she had to then move to the Battlestar Galactica community Laughter Have to earn those stripes again What if you then move to some other kind of fan you know media topic What happens when you move to the people who are talking about pharmaceuticals You need a tool Now the challenge think is that the tools that are out there were not designed for you They re really designed for software developers And I m hoping that I m going to anytime now advance maybe not There we go okay So the tools that are out there are really for software developers Even the graphical user interface versions of these tools are really for very very advanced users There is a tool and they re fine tools There s Easy Inet There is Gephi G E P H I it s actually developed by a group of francophones so it s Gephi but say Gephi And there s Pliek there s Gephi there s Easy Inet there are several There is in fact a Wikipedia page managed well founded by Jana Diesner who will be here on Friday to speak to you and she s got 40 or 50 tools listed So what do we need another one for Well Gephi has a motto and its motto is that it is the Photoshop of graphs We also have a motto And our motto is that we are MS paintographs Laughter And our goal is that we re the tool that you re
10. that you need to now stop and talk to somebody that you ve just encountered on the street what would you do You might think Oh will get out of their way The reality is you will stand stuck still right where you meet that person You will not move and you will instead of causing a blockage kind of human cholesterol you will actually be like a rock in the stream and they will just flow around you You would imagine that if you know there are 10 000 New Yorkers behind you So actually looking at what people do matters cause they do things that you wouldn t expect but the data collection mechanism here well was somewhat organic It was graduate students and graduate students are wonderful people but they have short duty cycles and they have this unfortunate habit of graduating if all goes well And so what we can now do is actually collect information about this off of your cellphone And in fact companies like Sense Networks are doing exactly this We are building a worldwide model of who is where and who is interacting with who at the petabyte scale and people like Sandy Pamplin and somebody is working with Sandy here there you are you know these are guys are building these kinds of models and they re not using graduate students at the data collection interface They re using them at the data analysis level And so we are at the cusp of a real change in the nature of social science research It used to be that a data set of 30 was lar
11. the cover think this is Alvin inaudible from Australian National University It s 1991 and this is a network visualization at that time was looking at BITNET was looking at BBSs was looking at message boards and looked at this image and said I would really like peppermint sticks but no want a picture like this That s what thought thought would really like to see my data in this way And you know any day now we re going to get there But it has been what That s oh my goodness Laughter Is it really It s 20 something years since then Yeah okay so it s 20 something years I m working on this Proof that obsessive compulsive behavior is really a good career developer Okay so our goal is to get to this to understand these phenomena and in part because now you know doing research the old way is now too hard This is the work of William H Whyte whose book City Rediscovering the Center is a great book also related to his shorter work The Social Life of Small Urban Spaces strongly recommend reading these books They re wonderful And in this image what we re seeing is data that was collected about where do people stop to talk to each other Where does an edge form Only it s a street corner in New York City in Manhattan It s where Saks Fifth Avenue is And it s worth noting that on these city streets 10 O00 New Yorkers will walk down the streets Now if told you that 10 000 New Yorkers are behind you and
12. will gt gt Okay gt gt Okay But so you know what we re playing with are ways of thinking outside the spaghetti bowl outside the bug splatter to say it s only a bug splatter because essentially you didn t take the camera and bring it into focus And so am a little frustrated by those who dismissed network visualization say you just have to finish the job that it s a matter of bringing the image into focus And once you do then there is some clarity maybe not on the gt gt think you need to refresh the graph on this gt gt Oh right okay Well maybe should do no don t seem to have control of it It s a slow virtual machine and we ve just done most of the computationally intensive thing and apologize But yeah so we are playing with these alternative ways of representing you know nothing says that every edge must be shown as an arc between A and D There are ways of aggregating them more glyphically representing them at a higher level And so we are looking at that as a way here we go will go to so that is the bundled edge see how they all got sort of swooshed down to a com So it s like what you ll see when good network engineers run cable right They bundle the cable very neatly They re very very precise about their zip ties So it is essentially the zip tied version of that but that s not what Ben is asking for he s looking for this Let s go here and go to edges go to other You want now where is t
13. 00 dollars you could probably get your data If you have 300 000 dollars you could get your data But if you have zero dollars you might have a problem Some members of the NodeXL community have been grandfathered into this data access tier where they can get more data and so we have built a service we call it the NodeXL Graph Gallery call it Flickr for graphs and email And in that gallery there are now a thousand graphs And as the user community for NodeXL grows it is our hope and aspiration that there will be a million graphs in the NodeXL Graph Gallery And that this will allow you to pretty much ask almost any question any topic and you might find that there s a graph in there already And if there isn t do encourage you to request the graph Go to our message boards ask if someone could get that graph and upload it for you And so we would then have addressed two out of the three problems open tools don t have a tool now you have the tool no data now you have data And so our next goal is open scholarship and of course we re doing events like this and others because we think it s really important to teach people not only the just nuts and bolts of using any tool but by learning those features of that tool really learn the larger set of network concepts to help you as I ll say later think link Think in the form of a network And when you do you ll walk out into the world and you ll never see it the same way again You will always b
14. That s digital media but it s not social media Social media involves people connecting to people All be it it is possible that they connect indirectly So added a Wiki page you added a Wiki page It doesn t mean that we added each other but we had a shared intermediate connection And so when you think about network theory and hope you do as you learn network theory it is a challenge and that there are many new concepts and these concepts tend to be interrelated that the meaning of one is actually related to the meaning of another and so until you ve learned them all you haven t really learned any But I will offer you the shortcut and that is that network theory has a lot to do with real estate You may recall the three most important things in real estate And that s good financing the foundation and available schools no That would be location location location And network theory doesn t quite have location location location cause location suggests that there is a north there is a south and there isn t in networks But there is in networks is position position position You want the prior image gt gt No gt gt Okay So what we re trying to do then is make it possible for me to achieve a goal that started to have in my mind in 1991 bought this off the news stand I m sure you all did too And you know buy Scientific American for the articles but like the pictures too And what really found compelling was just
15. These transcriptions may contain errors especially in spelling of names These are unfortunate and we regret that we do not have the resources to fix these errors Still we believe these transcripts will be valuable to many users Mapping social media spaces Marc Smith gt gt All right my dear friend my buddy Marc Smith PhD Sociology University of California UCLA and a co author with the book on NodexXL but just had great fun with Marc We ve been working together for five years in the NodeXL project Every Wednesday for an hour we meet on the phone with about six or seven other people and argue vehemently about what s the next feature to be added to NodeXL And we so when we get together we continue to argue vehemently but it s just wonderful gt gt Respectfully gt gt a dear buddy really love it And you should know it was his birthday this Sunday so we took him out to dinner too so we re having fun too Applause gt gt I m 47 laughter All right welcome back It s good to see you all here How many people just a quick show of hands were not here on Monday Wow Who was here on Monday Okay Why don t you guys go outside Laughter Okay cause I m going to do some of the same stuff It will be a little different but it s going to be somewhat similar on Monday afternoon for those who could come in early Alan and were here with Bernie Hogan and we did a much longer hands on demo and workshopy kind of thi
16. a developer And even then very few developers have used their toolkit to build data at the scale that we have so that we have now thousands In fact know half about a quarter of a million slices of graph from a variety of social media data sources And so we re learning things by looking in at microscope or into that telescope and things like the way that a lot of political discussion is pulverized Now it s probably not surprising to you at least for the Americans Maybe the Canadians who watch American TV they know were a little polarized in America just a little bit And you know I think we intuit this but can we measure this And so the image you see here is a discussion in Twitter amongst people who all said the word GOP Grand Old Party the Republicans and what do you see see two clusters with very sparse connection between them We see polarization We see a way of actually quantifying polarization as essentially graphs that had multiple clusters that had very little connection to each other versus sub communities which are essentially clustered graphs that have strong connections to each other but are still sub communities And so our goal is to both contribute to let s build a better tool and then let s apply that tool and do better social science Let s understand social media And by that mean how many different kinds of social media are there And I don t mean Twitter Facebook Flickr but within Twitter how many kinds of h
17. actually going to use not the one that you re thinking about learning And so we may or may not have achieved that goal but it is our intent to be the 80 percent solution for graph analysis Does that mean that we will have the feature that we need to do everything you re going to do in very sophisticated graph analysis There s a good chance that it isn t the case We re trying to be the 80 percent case tool However we do think that if you run off the edge of the NodeXL real estate when you then take the next step into the next tool you will have a much more motivated reason for being there rather than trying to climb a very steep learning curve So am trying to do the social media thing myself I m working as an independent consultant and also helping to lead the Social Media Research Foundation We are a 501 c3 not for profit California Corporation Our goal is to convert money into software know a lot of other people would like to turn software into money but we have a different goal And we will also then make that software free and open And so that s free as in speech free as in beer it s free The group of people that we ve gathered have to say is a motivation in and of itself There s some remarkably talented people who for one reason or another have been corrupted and been forced to engage in this project And they come from importantly not just a wide variety of institutions and time zones but they also come from a wide variety of d
18. ashtags are there How many kinds of users are there And can we taxonomize them Can we come up with categories Now when everybody is using the tool we do believe that we will accomplish a goal not unlike this project this is the Allen Very Large Telescope Array Paul Allen co founder of Microsoft has spent his money in interesting ways One of which is to build the world s largest array of radio telescopes It s in the New Mexico desert and it rather than trying to build a really really really big dish which structural engineers have said basically We re at the limit and can build a dish any bigger than the one they have in Puerto Rico they re not getting any bigger And the answer to that is Fine will take 1 400 small ones and they arranged them in this big cruciform out in the desert and that thing has the resolution power of essentially a multi mile wide dish Our inspiration then is to encourage people to also set up NodeXL and NodexL data collectors and collect data in all of the regions of the world that they are doing research on the topics that motivate you and then to share that data And in doing so create a triangulated map of social media to actually let us see all the different kinds of social media there are So all right social media has already in some ways peaked it was on the cover of Time magazine take some issue with this issue if you will Partially the issue for me is that English is ambiguous in this pr
19. ata section the Graph section the Visual Property section Analysis option Show Hide and Help And in the Data Import menu you can see just some of the sources of social media network data that we provide Twitter Flickr YouTube Facebook what else Flickr Email that s personal email or we also have an Add in to talk to exchange servers all these ways of getting data you can even download it from the Graph Gallery You could open it from your own workbooks maybe you already have data You have spreadsheet So long as your spreadsheet has person 1 or person 2 well not or and So basically if you have a spreadsheet with two columns with name and name you have a network Open that spreadsheet Open the NodeXL template in the same copy of Excel And then go to our Import from open workbook and you now have the ability to pull data in from your worksheet pull it in to NodeXL and analyze it But in this case use this the Twitter search network importer We have three Twitter importers with the various kinds of networks you can pull out of Twitter That s a User s Network You give me a name I ll tell you who follows them and who they follow and how those people follow each other if they do There is the List Network You give me a list of names or you give me the name of a Twitter list So if you re a Twitter user you may be familiar with the idea that in Twitter you can create a list of users If you give me the name of the list of users
20. co URL what do you get You ll get a bitly URL Laughter So you don t really want the bitly URL do you So we double unwrap and we take the bitly URL we also unwrap that We now actually have the URLs that we re actually mentioning You saw that just a few seconds ago in that list of top URLs Many of those URLs had been wrapped into a bitly URL and then wrapped into a tco URL we double unwrap them It does take a little bit of time So did this and it took a few minutes Mostly it did is to transit time of going to Twitter and bringing back all these data and then iterating all those names And of course you may have other challenges You may have the challenge that Twitter at some point is going to go it s enough for you No more for an hour And it could very well be that you can make about one of these graphs a day on one copy of NodeXL using Twitter You can make as many of them as you want using YouTube or for that matter even Facebook as long as you re getting the right kind of Facebook You cannot get anybody s Facebook ego network other than your own But you certainly can get a lot of the connections on a Facebook fan page very easily So we support fan pages and ego networks we don t yet support groups It s coming So we got a data set and that data set takes the form initially of just an edge list And here s the edge list And will sort this and do a little tweaking here and you ll see that what we have is a group of people and smile
21. down looking like a black and white image with no photos It was just the arcs The arcs didn t have thickness they didn t have color They didn t have all of the things that you might do to decorate a graph And so we have a tool in NodeXL called autofill columns and this allows you to map data attributes to display attributes And it is dynamic in the sense that if we go into the drop down you re going to see a row here for every column in your workbook in your edges worksheet you re going to see a bunch of columns If you add a column and in fact have a look at this let s see if can you see this column the add your own column column This is intended to convey to the user you know this is Excel you can add columns You can add formula you can sort You can filter You can do anything that you know how to do in Excel You still know how to do it only you re now doing it in a enhanced version of Excel And so you could add a column and it could be shoe size it could be IQ it could be income it could be zip code it could be area code it could be anything And you could then filter or decorate your graph on the basis of that data And so you would do that by coming to the menu going to autofill columns and you would say Hey you know want my edge width to be based on the edge weight and the edge weight was calculated by this thing that rolled up all the replies in all dimensions and counted them and then added that to every edge O
22. e a lot legal And that s true these people are very happy with you know a power of her but they re also talking about this word war on women Here up here this group they re very tightly tied to each other so maybe they re similar maybe they re different because at the hub this is all the people who essentially are around one news outlet these are the people who are talking to each other there s the Gates Foundation And so the algorithm decided that they were separate clusters but it also notes that it has a lot of connective tissue And here they re talking about contraception and birth control and the war on women too and repro health watch and P2 that s Progressives 2 0 insanity in women s health Okay but who is G4 Notice a few things about G4 very little connective tissue to the other groups They re not that connected and who are they They re people who don t like birth control This is it is these people who make this a controversy rather than it s like saying don t like aspirin mean why would you not want it It seems to me anyway maybe others have a different opinion but these are the people Teacup Pro Life Catholic Contraception Tea Party Pro Choice Abortion These are the people who for whatever reason believe that it s a really bad thing And they are the ones who make it a controversy by opposing it And now we can see and think made this point Monday afternoon that when you look at a bucket of tweets or
23. e thinking What are the patterns of connections amongst these people Who knows who How do they know each other Who s the bridge Who s the cutpoint Who s the hub And we re going to look for these patterns and I think once you start thinking in a network way you will have a very hard time avoiding seeing those patterns So in addition to the training we re also publishing and we ve been publishing think in two separate tracks One track is a CSS Ul UX kind of track where we are attempting to build tools and features that improve the way that people handle networks Not only the tools to get the data and calculate metrics about the data we actually have outsourced that to the very talented Dr Jure Leskovec whose SNAP library is our calculation engine In some ways you could argue that NodexXL is nothing more than a GUI over SNAP which is otherwise a C plus plus library But we think we re a little bit more than just a GUI over it because we ve also been working on innovations in the way that we represent the visualization of the graph So we have some goals in the project One of those goals is to address the number of lines of code USR user will have to master in order to use the tool and that goal is zero We ve reached that goal We have another goal and that is the number of human seconds necessary to process a graph before it crosses a threshold that we call good enough and that goal is zero Zero seconds of human involvement Once y
24. ect line There are that many others but every one of these is an edge It s a link It s a tie It s a connection And in aggregate they form networks And so our goal here is to encourage you to think link To think about look there all these entities in the world and they have relationships with each other and not all of those relationships are the same They come in flavors And in fact not all of the relationships even are of this type which is unidirectional that you know it goes from A to B We talked a little bit yesterday about how it is the case that some relationships are bidirectional and others are unidirectional We call them directed or undirected So directed edges are often not reciprocated So when you do something with someone that doesn t mean that they re going to do it with you followed you on Twitter it doesn t mean that you follow me Other relationships are reciprocated If I m you re friend on Facebook you must be my friend It can t be any other way So some relationships are balanced others are not I d like to note that you know some relationships you give the ride to somebody to the airport but they tend not to be the person to give youa ride to the airport You lend them money they tend not to be the person who lends you money But some relationships are inherently bidirectional and we talked yesterday about marriage right If you find yourself married to somebody who s not married to you Laughter It s probab
25. g to be mentioned in another cluster Well in a highly polarized graph the answer is none none So that means that no one in let s say the liberal cluster pointed at FOX News and nobody in the conservative cluster pointed at npr org None And this for me is actually a fairly concerning finding that we are talking about the same things and we don t even talk about it in the same way and that we don t have any common referent for that topic is a concern And so being able to analyze what is being talked about is a very useful addition to our analysis And then what happens if it s not a hashtag it s not a URL and it s not in that name then it s words and then we were giving you word pairs What are the words that have gone and some are social makes sense right It does make sense that those things would go together But if you are dipping into a domain that you are not familiar with getting this kind of a report can be very handy and I ll show you what that might look like in a live version So made the SSW map of right before started speaking And so this is that map and we are now in NodeXL and so how did get this data Look it s really it s getting bigger And TU2 3 4 and 5 all got mentioned but not 6 but wasn t talking at the time so maybe you ll tweet about TU6 So what did was to go to the NodeXL menu and you ll see that we have NXL of ribbon and like ribbons And in the ribbon we have a whole bunch of sections the D
26. ge It used to be that an observation window of a few hours was long It used to be that you know studying a region was a lot And now we are going to be studying the mobile connected user base of earth which believe is now about to crust 2 billion units And so we re going to have a real time model of where we all are After all who here and I think know how to ask this question who here does not have a phone on them Where did you leave it sir gt gt don t have a phone gt gt This is our outlier Laughter amp Inaudible Remark But you know a few weeks ago my wife left the house said you know Have a nice day went off to work 8 minutes later she s back why gt gt Left her phone at home gt gt She forgot her phone How many other things would you actually turn around and come back to the house for Your wallet your keys no you re in the car you couldn t have forgotten the keys So you know it s California we re all driving So you know this device has become essentially and there was an article recently Don t Call It a Phone was the article There s a ranked ordered list of which applications are used most often Do you know where a phone is on that list Fifth And if you ve ever tried to actually have a conversation on an iPhone you know why Laughter It s not a phone it s a web browser yes phone no And so these things the proposal was that s not a phone that s a tracker We should call them
27. graph a bunch of things and the relationships amongst them And our argument is that while the world clearly understands the value of HTML it will soon come to recognize the value of GraphML that we are all on the path to network consciousness an awareness that none of us are really isolates We re not individuals As a social scientist I m here to say we re not individuals We re actually deeply interdependent If we were individuals we d all speak our own private language And that s actually considered to be a pathology so don t do that So we want to make it so that in the same way the Firefox stepped up the game from you may remember links maybe you don t about the text version of a web browser before Mosaic there was links remember you know the sign of my great capacity for technology prognostication remember seeing links and thinking saw a URL for the very first time and said That ll never work Laughter Forget about it It s not going to happen And although it happened and so our goal is now to build this tool that lets you as non coders and as coders to communicate to non coders How do get a graph Where do we get it from Where do put it in once get it What do do with it once have it How do analyze it Where do I what those network analysis even entail And then what do do to visualize it And having visualized it how do find an insight into it and how do I publish that insight And that s the tool c
28. hain the data chain the flow of data and operations on that data that believe anyone who is engaging in network analysis has to perform And there s a challenge And the challenge is that pretty much any second year computer science graduate student can do all those operations in Python or Ruby or something And if say to the computer I sense that they were over on this side so if say to the computer scientist you know can you use Python to open a socket to communicate to an API to do a query to grab out a bunch of rows of data maybe a few hundred million rows of data and then put it into a scalable let s say MongoDB and maybe you re going to build a Dub Cluster If you re losing me that s my point laughter but these guys know And then you re going to take queries out of that database run them through other tools like igraph networkxX or SNA then take that data populate the database with the results and then take those results and then pass them to some other process to visualize them You don t have to go to school to learn all of that Laughter But wait a second you are going to school to learn something else and so there s an opportunity crossing You really can t do both Some people can and I it s a fine distinction If you can you should But if you don t feel like that s the path that you can walk down am here to say don t think can go there wrote my dissertation work in Pearl and then met real developers and they
29. he street It s what part of the house he was in It looks like he was in his front yard So here we ve gone from a tweet Inaudible Remark Oh maybe he was on the bus at that time Oh look at that Oh there he wave Laughter So partially this is to terrify you Laughter The good news is less than about 3 or 4 percent of all tweets have lat longs The bad news is all tweets have lat longs They re just it s just the case that the lat long stopped at the phone company and it didn t make its way all the way to Twitter So everything you ve done looks like this if you re in Leicester But laughter you probably it ll show you right here So we can do this with any set of tweets and then we will get from NodeXL these other worksheets We get lists of all the people and then this is stuff that was summarized in that caption but all note that we actually write that caption for you This is the graph summary tool press this button maybe you want to copy it to the clipboard you re going to write an email you re going to begin a report you now paste that into Word or Google docs and well it s not a finished report but you are about 60 percent of the way there now You now add that which only you can add which is inside story meaning and that s something that NodeXL is probably not going to do for you anytime soon So how did process this into this image Did it just come down looking like this hope not that s better The answer is no it came
30. his Tell me gt gt What is it gt gt don t know gt gt It s changed gt gt Oh dear how embarrassing gt gt Is that the thing where when you put them in the group reviews peer groups gt gt Yeah it s in the groups Go to the groups menu gt gt So you collapse all groups and then the edges will view the bundled graphs gt gt Yeah gt gt You don t need to collapse the groups just go the groups That s it gt gt I m not sure Ben gt gt No no no no no no gt gt What would you like me to do gt gt Do not do it gt gt I m lost gt gt Okay gt gt All right I ve got to rehearse that gt gt We ll talk about that later Okay we know our tool I m telling you It does networks now gt gt What gt gt It does networks now That s really cool gt gt Didn t you guys literally write a book on this gt gt Yes Laughter gt gt We should find the book gt gt know I ve heard that gt gt The book is on version 113 We re up to 221 so if you want to help us write the second edition these features I m talking about like the Group In a Box are newer than the book so they re not in the book And we have to go back in this that s yet another job mean it s one of these wonderful things where there s a great audience and great interest within those powerful tools But it takes a lot more energy than we all have when we have students who are after a PhD
31. ho are algorithmic will decide to contribute to and use to help deliver their results to the rest of us For the rest of us think it s the tool that gives us the super powers of a second year computer science graduate student without actually matriculating It s grad student in a box Laughter And so our goal is to be that Firefox of GraphML the thing that let s you think link with a much lower overhead a shallower learning curve And hope that you will also see that it is possible to generate publishable findings and results by bringing and there you go That was it gt gt So what you re seeing now well we ve got to set kind of light but it reduced all the connections between to this thin gray line here Here we go It s coming a little better So you re getting so we ve now and it gave more space over here We probably want to resize this a little so we can see it gt gt No yeah it s going to be a inaudible gt gt Oh yeah gt gt Okay do it gt gt Oh dear too late gt gt Too late gt gt Oh dear In any case gt gt You see that line was back there Laughter So how do you progress if you wanted to join us Ask your questions on the message board Send your mail to me or Ben Catch us while were here for the rest of the week if you have a data set Somebody gave me a data set I ve got your graph And you know we d like to help you get to social media insight or network insight broadly but certain
32. ia to the US senate So there is a book there was another book a fine quality book with my now sadly departed adviser Peter Kollock And we have applied these tools to a variety of topics And this one was used as a featured example in a recent publication at a website which was Noshir Contractors conference he hosted it at Northwestern in early June And Jana Diesner ran a workshop on Words and Networks how do you use the words in networks and the structure of networks to inform our understanding of a graph in general but a social media graph in particular So this is people talking about what surprises me to be a controversial topic It s the word contraception or contraceptives and it turns out that these two groups are strong positives on contraception They talk about contraception in case you missed it birth control Griswold Anybody know what Griswold is Yeah Griswold versus Connecticut 45 years ago In fact this day these days in June this was the 45th anniversary of Griswold versus Connecticut What was Griswold That was the court case that went to the Supreme Court that essentially made contraception legal in the United States of America So it s hard to imagine here folks imagine a time it s illegal Not only was it illegal it was illegal to talk about it in the US male it was illegal to share information about women s health reproductive health Okay but that 45 years ago you would think Hey things are good now right We hav
33. ial It s about the exchange of objects between people turned out that this network thing was not about the efficient use of super computing resources it was about exchanging photos and music and text messages and URLs And essentially every time you do one of those things you form a connection And in aggregate those are collections of connections and collections of connections are networks And of course the other big issue for us and opportunity is that humans have been social prior to social networking services They were actually social before electricity So humans have been doing this social thing for a very very long time But throughout all of human history being social has been an ephemeral phenomena But with few exceptions most of the time when you talk to somebody and walk away the conversation has evaporated and if no one saw it happen it might as well had not happened except for the memories of the people who participated and those memories are flawed and you can t quite recall who you talked to yesterday You know one of the greats in social science studies is to simply ask people what they had for breakfast yesterday and most people can t tell you Maybe breakfast is more stable but lunch they can t remember We don t remember these things And so on the net the great opportunity is that silicon even in its most raw form is really good at storing the foot prints And the difference between silicon and well unprocessed and
34. isciplines It s not just computer scientists love computer scientists really do but it s not just computer scientists So by bringing social sciences and user interface designers and computer scientists together what we ve done is created a tool that meets the needs of users rather than developers of that tool So we have a goal Money converts to software and we want to become not unlike Firefox free and open tool to browse a really important data structure But we re different from Firefox because Firefox is not a web browser and NodeXL is How could that be Well Firefox is not a web browser Firefox is a page browser You ll never see webs in Firefox You see pages You see a node You see the vertex It renders the content of a vertex but it never shows you the associations between multiple pages It can t It s a page browser NodeXL is a web browser Okay now give up because we re not going to win that battle Obviously yes Firefox is a web browser and no we are not one Maybe we are a net browser I ll take that So what we want to do is be like Firefox but for some other data structure not HTML and this other data structure is known as GraphML And if you Google up GraphML you ll find at graphml graphdrawing org that there is this really nice simple clean open basic XML schema that is now widely being used for a variety of tools to interoperate and what is it It s the file format for storing a collection of connections a
35. ital tool and literate and computer savvy find that indeed that it needs training and found it to be very much different today well inaudible left that was 60 percent Obama 40 percent inaudible So just figured you know which kind is always represented these gt gt trust the IOM market more than do this have to say At least because I m going to identify this I think fatal methodological flow says who right If you simply get a lot more pro people to be more pro that s not news and if you got anti people to be more anti that s not news either So let me see if can get some of the other comments here ma am gt gt Okay cause the thing is is that in terms of elections in America how much would we make does matter How much you want to spend on political eyes where you target them and when you target them doesn t necessarily change anyone s mind it gets buds out of season for election So what question you re really asking is does Twitter matter in the same way that other guys at local advertisings do So Twitter does matter and you know we re going to make 50 000 more tweets in the right places and the right you know anytime when people want to do or they may in fact have actual political impact gt gt Okay gt gt And don t know if the answer don t know if Twitter s being not actually inaudible possible or hurting people but making noise that s always been made by making it louder in the right time
36. looked at my code and went And you know maybe it was a good experience but think maybe the best experience is to learn enough to know how to talk to a talented software developer and then find one in partner Or we have a tool for you So we are trying to build this tool and we re trying to fulfill the foundations motto open tools open data open scholarship And we think that these are three goals that are really important for us as a culture as we wholesale move human societies into cyberspace If you were an ethnographer and you wanted to understand the civic engagement you might go to the public square and watch people talk to each other and it was a physical space and you could occupy it with a notepad and a pen you could do good work But the civic square is now in cyberspace So no matter how many people actually were in the Tahrir Square and there are many Tahrir Squares so even if we added up all the people who ever were in all of the Tahrir Squares more people talked about those people in Tahrir Square in the cyber Tahrirs of the world What is the magnitude for it Millions of us talked about Tahrir hundreds of thousands of people were in those Tahrirs And so how do we now as social scientists as people interested in what s happening to human cultures in an age of technology what do we do How do we get our data and do some meaningful analysis of it And so open tools open data open scholarship our goal then is first to open t
37. ly social media insight with a lot less effort And we don t assume that giving that it s a reduced amount of effort that then you know it s time to watch Breaking Bad for 14 hours It s then you know time to spend that time Instead telling the story of your graph or maybe repeating the same analysis over a hundred time slices of your graph which becomes practical because you can now automate the collection of this data You can as a matter of fact maybe could show you very quickly So yeah here I ve got a virtual machine up in the Amazon Cloud and that is what NodeXL looks like when it wakes up using task scheduler every 15 minutes it wakes up and says it s time to do the Romney graph It s time to do the Webshop graph It s time to do the you name a topic that s relevant to your research graph you can schedule this It s going to do that and when it finishes it s going to write a file to the disc and that s going to be the set of connections amongst all these people Then it s going to open automatically a copy of NodeXL And then it s going to hand that data to NodeXL and it s going to say do the thing you do with it Do that recipe thing And then it s going to generate a report put an image on my hard disc and I m going to see you ve got a directory filled with these images not unlike the front page of graph gallery And then you are able to say what s new What s happening What clusters have formed Which ones disappeared What s going o
38. ly good to find counseling But other relationships are that way too If live near you you cannot live far from me right Unless there was one way streets inaudible but it s really not possible And so relationships come in flavors there s unidirectional bidirectional they come in types there s friend follow favorite all of those kinds of things And this is a good thing for us because we now live in a world of and this of course is a very abbreviated was there re all of these different kinds of social media And theoretically the challenge is how do we make sense of all the diversity that s out there on the internet And there is good news Despite the fact that blogs are not Wikis and Wikis are not Facebook and Facebook are not Flickr and Flickr is not YouTube although it is sort of like YouTube only the pictures don t move but for the most part these different services They all have one thing in common and that is that they all encode networks Your email has a reply graph your Facebook has a friendship graph and a comment graph and a like graph The universal data structure that underpins the entire social web is a network And if you are in a piece of software that calls itself social software or social media and there is no way for people to connect the people it s not social media So I ve heard many marketers talk about Oh you get our app and it s social media cause you can look at our brochure and that s not social media
39. n In fact the study of collective action often talks about the tragedy of collective action that there is a failure to accomplish successful collective action And the studies of failure usually come down to one thing Why do some groups succeed and why do some groups fail and the answer is information about other people s choices And so the internet has made sharing information about other people s behavior so much easier that we are now in a situation where it is possible to coordinate your behavior with others other people who aren t in your time zone who may or may not actually speak your language It allows for the aggregation of a critical mass of a labor And when that happens Wikipedias are born operating systems are created browsers are written message ports are populated with useful questions and answers lots of really nifty resources We talked about Flickr earlier today One of my favorite things perhaps you will help me by collectively adding your name to your photo when you go to my photo screen on Flickr So how do these things work And how do we study them How do we study them in a actually empirical quantifiable objective robust way For that matter even if those weren t our goals how do we cover these topics with any sense of scale don t know about you but started in this business by reading messages And there is a point when you realize you will never read enough messages It s not possible or maybe you have mastered
40. n in my various topics of interest And you re now doing this at 60 000 feed with the ability to have of course dive right down into that guy s living room in England So contacts and detail and overview and in a package that we hope will help you achieve your research goals And that think is our story gt gt Woo Applause Questions Yes you do inaudible gt gt heard Marc many times It s always great fun and it s wonderful Any questions or things to solve Yeah let s take a few and then we re going to sort of get gt gt There s crabs and beer between us and crabs and beer Yes sir take your shot gt gt Do you know anything about what the general with the political gt gt Ah yes divided Laughter and Inaudible Remark You know would say that the left and the right both very aggressively and actively use Twitter And have a tool that allows me to look at individual spaces in Twitter You would have to go to the Twitter people and say okay you know given a 130 million tweets per day what s the breakdown can t answer that question don t have 130 million tweets a day What do have is the ability to take a picture of today s set of tweets on a topic And when look at you know topics that guess would be right wing topics they are plenty populated with active discussion When look at left wing topics plenty populated with lots of discussion And when look at topics that are at the boundar
41. n in the graphs which we ve looked at there s the Republican side is more closely woven together than the Democrat side And that don t know about the magnitude but the woven this tight gt gt always find that should what Libby have the next comment gt gt Yes right gt gt will answer each of those questions while inaudible gt gt All right laughs gt gt Every single one gt gt Good advertising gt gt Congress says replies have had 530 tweets since Mark said 15 number tweets so in the last 45 minutes Congress generally had 6 000 tweets a day So Jenny studied from three years ago gt gt Yeah gt gt While useful is gt gt Very old right gt gt Yeah gt gt But you may have gt gt It doesn t capture a day in gt gt Maybe doing gt gt Perfect teaser perfect teaser tomorrow Libby gt gt Sorry we ll look Congress not the public Congress tomorrow gt gt Okay yeah So these are very different This is people talking about politics not politicians telling you about politics But will underscore that point Whenever see a polarized discussion birth control contraception and l m going to find that there is a group and it s isolated from the other groups and it danced And those are conservatives And they also lack any mass media outlets in their group Whereas and I say the color is that what we find in the liberal groups and you know they re not Liberal
42. ng a tutorial on this thing NodeXL And we re going to move a little bit more quickly you know in a higher rate but the main theme here is that we have designed a tool for the non algorithmic so another show of hands please How many coders are in the room Okay You guys can leave laughter All right well my argument is that while the coders may be a little roll their own they often and seem to they gt gt already suggest that we smoke it Laughter gt gt If you can figure out how Laughter Just tell me how it goes That what gt gt You ve heard of smoking code Laughter gt gt That what we re trying to do is develop a tool for the non programmers and maybe for the programmers who have to talk to non programmers and would argue that the distribution of programmer to non programmer on Earth runs about 99 9 percent non coder to about 0 01 percent coder And so coders tend to come together homo folly remember homo folly Laughter And so they often don t realize how non coding the rest of us are And have often been told well why don t you just learn to code And we ve had some discussions about that And one of the answers is that I m really bad at it and don t know about you but you may not be all that good at it either had a son who is a musician He has been spending seven years learning how to play the saxophone I ve watched him And so know that you blow in one part of this thing you run your finger
43. o these are the top hashtags in the entire graph SSW12 not surprising And then what do we have Tu1 that was the Tuesday 1 session It looks to me like the well it depends on the time of the graph so I ll have to make another graph but maybe the Tuesday 6 is not getting tweeted about tweet quickly Help yourself Laughter HCIL inaudible survival ASA NodeXL all of these are the topics of discussion right now the top topics of your discussion but not exclusively In each of the subgroups you may be talking about slightly different things And in fact we have applied these kinds of techniques to the analysis of politically charged to polarized topics And we ve created some interesting findings as a result shouldn t steal the thunder of our final speaker on Friday Professor Itai Himelboim from the University of Georgia Department of Communications and Journalism But working with Itai and using NodeXL we ve been studying highly polarized topics in Twitter Topics like contraception or GOP or Medicare there are lots of them and we get the two big groups kind of thing and then we ask a question What is the rate of overlap between the use of hashtags words and URLs We re both talking about the same thing do we point out into the world in the same way So what percentage of overlap do you expect there to be in some of these graphs That s a this is a question of how many URLs that are mentioned in one group or one cluster are also goin
44. of like you really like you feel socially obligated to link you I m into you know you you know it could be that I you know we met at a conference it seemed like the thing to do you know There are lots of meanings and nuances And so are you my friend Yes no doesn t quite cut it So what we do know though is that while there is this preponderance of weak ties there is some strength in all those weak ties And so we ll invoke Mark Granovetter seminal article The Strength of Weak Ties and he asked the question where do you get a job Hypothesis you get a job from the people who know you best your strong ties Answer No incorrect you get job leads from people who don t know you all that well Why would they keep you job leads Why would they be the source of information They don t know you that well Well because they don t know you that well they know a lot of people and things that you don t know They have news They are from a distant part of the network This is by the way why we go to conferences This is what conferences are for to sustain weak ties and why do you go Because that s where your next job is coming from It s where your next collaboration that s going to get funded by the NSF because you need that diversity It s why we go to conferences It s why we go to events to meet new people because or reestablish connections with people we ve met before who are acquaintances cause they are the source of a great deal
45. onoun and that it could be in the individual right it was individual and plural Now was born in New York and raised in Philadelphia we speak a dialect And you go to the a shout out to the Drexel people Go Dragons And in Philadelphia when you say you you mean one and when you mean to say more than one you mean that would be youse laughter Now if Time had actually put on the cover yous then I d had no problem with it because what it suggests to me is that when you say you what you mean is the individual that the internet was really about people and it is but not individual people It s about youse or if you re Southern that would be guess y all It s incorrect was corrected was in the South and was told that y all is of course singular It s all y all All y all is plural Laughter So all y all is what mean about the phenomena it s not about individual It is about prominent people within a community but it s not communities it s about collective action It s about more than one person It is true that often there is a very small group of the maximal contributors and we look at that you know you ll see millions of editors in Wikipedia but you really see on the low order of tens of thousands who are the core of Wikipedia That s true But even those very few editors are not alone and they form a community And so it is about the collective that we are really most interested So social media is inherently well soc
46. ools and we have built a tool And NodeXL is that tool it s the network overview discovery and exploration add in for XL NodeXL Someone once told us that acronyms cannot be trademark and we can be sued if we did that We believed them so we are NodeXL We were something else before and we got sued so we re not that anymore Laughter We re really sorry So NodexXL is this tool you get it you plug it into XL and yes there are caveats It doesn t run on a Mac It s caveat have a Mac but it runs on my Mac and that s because I m running Windows on the Mac but you know you can use a virtual machine There aren t problems and we d love to run on a Mac and to the software developers in the room Laughter We should talk And of course we re really thinking that we re just going to skip the Mac and we re going straight to the web and so we should talk because there s a lot of overhead in getting to the web and leaving our good foundation XL behind So we built this tool we believe it is a friendly end user tool It allows you to do many of the very sophisticated kinds of analysis that you ve seen this morning where there were networks and metrics and all sorts of stuff going on and maybe you felt a little left out Maybe you looked at not that side of the room that side of the room maybe you felt like Wow that s really cool but can t play in that space And so we want to say to you you know there was a time where the brochure the new
47. ou d kicked off a graph we believe that you can get a graph that is good enough in zero human seconds of investment Now does that mean that your work is therefore done We have a joke that there s going to be a new button in NodeXL called write paper There s another button after that called publish paper and a button right next to that called get tenure Laughter But no We believe that once we ve given you these zero human seconds of involvement work product then that s where your work begins This is where the storytelling begins This is where the insight begins because the machine can count things but it doesn t know how to tell a story If you watched the Hollywood movie written recently you know that that s true They don t know how to tell stories believe that technically all of the movies actually must have been written by machines So an example of this so track one actually tried to build better software to make it easier to do this Track two use the software and think track two is really the social science track And that is apply it we built a microscope what can you see We have a telescope what can you see Now once you say the word telescope you invite comparisons to Galileo We are not Galileo but we are definitely seeing things that have not been seen before Whether or not these are in fact the moons of Jupiter don t know but we are seeing patterns in the data that were very difficult to get to unless you were
48. present in the graph rather than external forces coming into the graph actually kind of like having external data coming into the graph because this allows me to see that for example not so much in this graph but in some graphs you ll find a very large image indicating somebody with many followers And yet they are very peripheral in the graph And like to refer to these people as visiting dignitaries If you are so lucky it has to have Oprah to read about your topic that s great and no doubt at all it will attract an enormous amount of attention But that doesn t mean that it will represent the center of your graph She probably isn t And so people with a lot of followers are not necessarily the people with the influence within a bounded community within some space of discussion And so we can then use the metric tools here Let s see if can make that scroll There we go And we provide a whole bunch of metadata about the graph We re going to tell you things like here s the number of vertices there s 52 people on this graph How many connections are there What s the density of the graph Graph density is very high it s 0 19 That s really high We re also going to tell you who the top people are That would be the Webshop account Ben Jenna Jenna there you are You are among the most central people in our graph And then that would be me then PJ Blessy inaudible there you are In the top ten Mediamum phonetic hello there you a
49. processed silicon is in eight hours the tide is going to come in and wash this all away With digital footprints there is no tide Nothing is going to erase the digital footprint at this point thought it was interesting that it was now newsworthy that Facebook has now committed to the ability of users to actually delete a photograph That s news You can delete something And they promised they ll get it done in 45 days or less Laughter And I m not too sure that they can even guarantee it When you say delete what you meant was don t show it to me anymore make it less findable but you know there s a back up tape You know that there s another drive array somewhere with these bits And so bits may be more immortal than humans think there s a kind of law like regularity The only bits you cannot find or the ones you really need right now the only bits you cannot get rid of are the ones that are most embarrassing to you So patterns are left behind and these patterns are the patterns of ties and there are lots of kinds of ties And these ties are all of the internet verbs These are the things you do when you re clicking when you re dragging and dropping when you re hitting that carriage return what have you done You liked you linked you replied you reviewed can t read let s say the list like link reply rate review favorite friend follow forward edit tag comment check in think send is missing Maybe we should add send a dir
50. r and believe the answer is no The number of strong ties in our networks is not growing It s about same And Barry very poetically says that the number of strong ties you have is about the same number of people as you have chairs that could be placed around a dinner table on a holiday night and you put the leaf in That s how many people you know really well The people that you will send a birthday card to call on a weekly basis you know when they re ill you go and visit them or you call them these are your strong ties And the internet has been criticized because it is now seen as the land of weak ties And yet you know what the punch line is it s true that not all 750 people that have on my Facebook list are really my friends And appreciate even you know two days ago was my birthday and was really pleased got all these happy birthday greetings but don t really know all those people And it was nice and said thank you and think you should like and comment and say thank you to every single birthday greeting It s just a nice thing to do but all of those ties we re not really strong ties In fact the issue of these ties are very ambiguous In fact technologists and we re going to that side of the room now technologists have taken the most analog thing on earth human relationships And they ve turned it into a binary Oh yeah you know Laughter Are you my friend Yes or no But it could be lots of things like you kind
51. r we might come over here and go to the vertices tab and on the vertices tab it s going to say Well what drives the size of the vertex And so PJ this is going to be for you here we go This is vertex size it s currently set to followers that s my default preference we re going to change it to between the center alley And then I m going to hit the autofill button And we re now going to see that the graph is going to it s going to populate all the data and it s going to re render the graph and the graph is now going to use size And the data that drives the size indicator is not going to be how many followers you have It s going to be essentially a reflection of your position within the graph It s going to tell you whether or not you are essentially Jenny or Jenny Or Jenny laughter or PJ gt gt Yeah that fixes everything gt gt think that does yeah So where are you I m sorry gt gt Over on the right there gt gt Oh yeah there you are I m sorry with the hair That s right yeah okay So there you go and there s his tweet and notice that we have a little tooltip there And so we ve changed the size and was that easy have to tell you I m on the support list for all the other very very fine quality network tools that are out there and one of the most recurrent questions asked by users in the support for are how do change the size or the color of the node I m just going to argue with those guys Gentlemen in
52. re and Hussein hello sir Jen Preece That s Jenny You made to top ten And Jenny Corn phonetic there you are Yeah so guys have more than the others not tweeted a lot Not had more followers than the other but connected to each other more So these are our connectors There are hubs And so we can say something about the nature of our internal community here and I ve been making these maps before during I ll even do them after and of course we have theoretical reasons imagine that we will see densification of the graph over time in part because edges don t die And so any edge that s added will always be there Unless Twitter stops giving us data about last week in which case some edges will disappear after a while So we will also tell you who got replied to a lot but not just in the entire graph but in each of those clusters So you might not be the king or the mayor of the whole hashtag but you may be the mayor or the most central person in one of the subgroups We also we ll analyze the tweets So these are the top tweets our schedule the video the graph gallery some tweets an image more of the Webshop links so these are all of the links that are being passed around in those tweets and they re ranked by frequency I ve mentioned And then we break them out by groups so that you could see the different subgroups are actually tweeting on different topics pointing to different resources We also then tell you about the top hashtags S
53. rty laughter Soecter was immediately voted out of office at well 2009 he lost his job But in I m sorry Inaudible Remark Maybe mean okay we ll discuss that believe it was Collins who just retired or it s Snow gt gt Snow gt gt Snow just returned Collins is the last one standing So people often ask is between the centrality good It all depends Laughter like to say between the centrality often means that you have arrows in the back from both sides Laughter gt gt But wait a minute the irony is that gt gt There s the punch line gt gt Okay gt gt There s the punch line this was 2007 What happened in 2009 Some other other guy became the president and he had this idea that hey in really tough economic times maybe it s the role of government to assist the people of its own nation It s a radical idea appreciate that But only three Republicans voted for the Obama stimulus Who were they Specter Snow Collins but this was two years prior to that So in 2009 we could potentially have predicted who of the Republicans would go with the Democrats and vote for the stimulus Well you might calculate the voting record graph and we would see that Specter Snow and Collins are the only three that are likely And that two out of three are now gone So there is some value to looking at the network structure of all sorts of things not just social media You could consider this a kind of antisocial med
54. s you know you have so many talents You re so good at so many things why should we ask you to be good at everything And so this is another example of the don t run with scissors approach We take the isolates away and we do something simple That s known as grid very computationally intensive Every you know if we just grid the isolates and we order the isolates by some size attribute and then we grid them and what you get is sort of kind of a histogram or at least a gridogram And then we box we Group In a Box these things and you ll find that each of these clusters it actually turns out that there was a hub and a spoke kind of structure there and that was not visible when we saw it in this form We really couldn t disambiguate the clusters Now we can disambiguate the clusters and this took zero human seconds for that transformation It took some compute seconds you just saw that but it does take zero human seconds Cody has just helped us implement network motif simplification This is an example from the book This is Robert Ackland s Web Graph built with NodeXL with data from VOSON the Virtual Observatory for the Study of Online Networks It s a service from Professor Robert Ackland at the Australian National University in Canberra And VOSON is essentially a web graph crawler and it brings you back you give it either a bunch of URLs and it finds out they re connected to each other And then spiders out from there to some configured radiu
55. s a group of people in any democracy who will vote for the person they think will win making this very dangerous So I ll be right with you but got to go back there gt gt know because afterwards they say they voted for the winner The number of people would say they voted for the winner after an exit poll jumps by 10 percent gt gt voted for Mondale Okay be right with you I m going to start back there and we ll come around think we ve touched a nerve ma am gt gt Why don t we decide before by chance and was trying to find I think what is not involved is it s just has one Inaudible Remark gt gt And second was trying to find a book that say inaudible because on some days it s like 72 to 20 what does that number mean gt gt Put it this way If gt gt And maybe you d know because you would look that it s more left but couldn t find out what does the numbers mean gt gt So think there s what Jenny may critic I m not that sure believe it s telling you percentage of messages about from Twitter that are more negative than the set of messages about Obama or Obama mentioning tweets as compared to all other tweets I ve read the articles from Topsy don t Inaudible Remark Not only that but it also depends on some magic pixie dust called sentiment analysis gt gt Second was looking at the IBM the IOWA and the kind of market inaudible office enterprise which also has dig
56. s folks they are right of center right mean nobody here is actually a laughter or advocating the nationalization of anything Let s just you know put this in the right context There is no Liberals on these maps But what we find is that the so called Liberals they have a lot lower density and they have a lot more different hubs And so what like to say is that the problem with the left is that it tolerates diversity
57. s or you give it a search term and it goes to Yahoo Search and then it gets a bunch of URLs and finds some connections amongst those And you could use this for all sorts of well motivated research questions like how do the banks in Lebanon all connect to each other How did their web sites connect to each other Well there are only what 18 banks in Lebanon all of which are looking for money right now Apparently we somebody gave them a virus can t imagine what that would be and you know it would be interesting to look at the patterns of connections amongst all that Or maybe regulatory agencies or maybe it s a different sector of the economy maybe it s military maybe you can do all sorts of things with how do websites connect to websites and we come out within like that And we like these graphs these are pretty graphs and a lot of graphs have this sort of there s a hub or a network head and then all of these sort of flanges coming off Those are single tips those are people who only have one connection and in this case it s a webpage that only connects to that hub but doesn t connect to any of the others So this is let s say 10 sites all connect to one site but those 10 sites do not connect to each other What Cody has done he said Well why don t we just remove all of that And this is what it looks like when you ve simplified the graph It s a way of visualizing with a lot of the clutter gone And so we think that with development this
58. s that each cluster that would otherwise impact the other clusters that would maybe they would interpenetrate they would in some ways it would be like when galaxies collide right And there s your sun goes spinning out of orbit or the moon moves out of orbit and then we have a very long TV show about living on the moon and traveling through space But what we ve done is we ve isolated each cluster we lay it out and it turns out it s better Not just because we say so but because Cody Dunne says so and who s Cody Cody is a graduate student of Ben and he s been building tools that measure the fitness of a graph that tell us how good is your graph or more to the point how awful is your graph Because it is the case that people have criticized network visualization words like ball of spaghetti bug splatter hair ball These are all the words used pejoratively to say network visualization what is it good for Absolutely nothing say it again But it s not true bad network visualization is good for nothing but with just a few techniques and there s more to come And hopefully over the next day or so we ll show you what Cody has up his sleeve something that our team calls network motif simplification in which a lot of the clutter of graphs are reduced by simplifying certain motifs or common patterns that recur in networks We take them we remove them we put a glyph that says I used to be 14 people who all hung off of one guy Something we called
59. s up and down on the buttons That doesn t make me a saxophone player And talking to a lot of Python developers and knowing that you write words on the line and hit return doesn t make me a software developer either But I m a social scientist and I m a geeky social scientist I m really interested in what is happening online call this stuff computer mediated collective action am interested in collective action Collective action is the core concern of the social sciences would argue anyway that sociology certainly is mostly concerned with what happened when two or more people get together and do something that would either be very difficult or almost impossible for people to do on their own And so it takes two to tango and it takes three or four people to move a couch It takes groups of people to accomplish things And when we look at the internet what we see is an enormous domain of collective action We just heard all about Wikipedia and what is Wikipedia made out of It s made out of people It s made out of not just one person It s not that the person some you know Mr Wiki wrote the Wikipedia It s about well think the answer is it s about 100 000 active contributors and about a million and a half less active more casual contributors And so the internet seems to be about harnessing the swarm bringing collective action into focus and getting over all the obstacles to collective action and there are many obstacles to collective actio
60. sletter the short novel the report it also was something that was probably not within your school skill set It was also something that a professional was necessary You had to go to an illustrator you had to go to a printer and printers didn t do just you know here is the car key go to the Xerox print what you like They actually did what you do now in front of Illustrator Word or any of the other tools that you re going to use to put together some text And so what happened to illustrators They moved to higher ground They do the fancier stuff now So can t software follow the same path for networks would argue it can that we are the desktop publishing of networks that we are the tool that pretty much anybody can use and that gives the 80 percent solution to a very large population There are certainly things that you can t do but maybe they aren t the things that you need to do So we ve also been working on this concept of open data Many people as we they get into the space of social media research discover getting the raw material of that research is not easy Twitter has things called rate limits and API call budgets Facebook has permission requirements All of them require some kind of authentication and many of them will limit the amount of information they will give you Worse services like Twitter are happy to give large amounts of information to anybody but you so long as they have a big fat checkbook And so if you have 60 0
61. stead of showing them combine them gt gt Oh gt gt So the beauty of this is this graph is not a great example Now we refresh gt gt Yeah it s doing its thing gt gt The point is it just strips away of lot of the extra clutter and makes for a much cleaner graph In this case there are only two major groups that we re only going to see one you know major edge between them But it allow it hides what think is a lot of the clutter then you can actually read the names on the labels Why is this taking its time gt gt 1 gig virtual machine running Windows on the Mac gt gt Right okay gt gt Not the ideal demo machine gt gt Okay mean this sort of popped up And it just sort of just read at the graph here and it just really cleans up So we can see all the names Nancy Baym is the biggest between some inaudible Jenny Corn right over here PJ Rey myself and Marc Smith and then Webshop UMD just lead the way Come on you can do it gt gt Okay so hope that this will have given you now I m going to leave it it s going to come back but I m going to do you know closing It s time for graphs gt gt Yeah yeah yeah it s time laughs gt gt And beer and graphs gt gt Right gt gt And anybody who wants to hit crack the crabs and I ll you can open them and I ll eat them gt gt It s a lot of work gt gt So it s the tool for the rest of us It s a tool that hope those w
62. t think that s a CNN outlet So there is mass media in this cluster there are no mass media outlets in these clusters So this is the group of defenders with if you use the word Obama Care there s a really good chance that you don t like it However if you were to use the word ACA there we go and that of course is the Affordable Care Act What you get is almost no one who doesn t like it right So here if you re going to use the word ACA you re not even really going to find let me get you re not going to find a cluster that has the word TCOT as its highest frequency hashtag And so just the words you used to talk about the phenomena is going to determine what company you keep and depending on the word you use you may find yourself in enemy territory or among right thinking people who share your view whatever that view maybe And so how you measure this is really fraud It s dangerous to say Hey send me on ACAs it s pretty much positive So with that let me just now gt gt found it gt gt You found what gt gt The combined edges Do you want to go back here gt gt Okay laughter persistence gt gt wanted to gt gt Persistence and composure gt gt We thought of a lot of arguments gt gt All right so yeah where is it gt gt It s under layout gt gt Oh yeah gt gt Go to layout gt gt The hidden layout option it s in there gt gt And then over there into group edges hit in
63. t their adviser is the Almighty laughter And that therefore it is it s a category violation How could there be another And so this kind of dies out for a while but these original graphs these hand drawn graphs really are the origin point of social network analysis This was the map of American Football Team and who liked or disliked other people on the team This was an early map of people in the Western Electric Wiring room a manufacturing plant outside of Chicago where in the old days humans made phone companies switch devices physical mechanical relay devices And these were who worked with who who helped who in that environment And so these very prude drawings are really the origin of network visualization and we re now on a path where we want to make network visualization the visualization of social media data a lot easier And would argue that we are still in 1959 It was only what 40 52 years ago that we took our first ever photograph of the planet we live on So that was 1960 Could you imagine if you are a meteorologist in 1959 You have never seen you planet You have never seen a gestalt image of the entire phenomena That is what internet research is today And unfortunately next year is 1960 next year is probably like 1923 we are long way away from seeing a holistic view of the whole internet of the whole social web And so what I m going to suggest our partial fragmentary steps towards that goal So here for example
64. tion known as Treemap and network visualizations And of course Treemaps were designed and developed here at the University of Maryland Ben who did that work gt gt think it was me gt gt Yeah it was you Yes so Ben invented Treemaps in the early 90s and initially you know to apply to things like how do reclaim a wasted file storage on what were then limited hard drives You had what hard drives with what 5k back then or 10k gt gt There was this megabyte Mac in our lab that was shared by 14 users And couldn t figure out who has bulging space So had this concept and it was build by graduate student Brian Johnson and the rest is history it s just on very long gt gt And so you ll find Treemaps in lots of places smartmoney com has something called the Map of the Market The Hive Group sells a version of Treemap you ll find Treemaps in a lot places There s now a Treemap for d3 js there s a lot of Treemap out there And anytime you want to see a complex hierarchy in a single page you probably want a Treemap What we did was we sort of we adapted Treemap we created a single level Treemap and made each group or cluster in a network we gave it a region proportionate to its population So how many vertices you ve got how many nodes are in your group that determines how much real estate you get and then we layout each cluster in its region as if there was no other part of the graph in splendid isolation That mean
65. trackers There is that This thing knows where am it knows who l m with It knows when landed It knows where take off from It knows you know pretty much knows my entire pattern of life In fact believe it was your group who was it You ll clarify this Research team found that you don t need to have somebody s social security number to have a grid for them a unique identifier for them All you really need to know is the GPS coordinates of their phone at 2 a m and 2 p m cause there are rarely two people on earth who have those same coordinates Maybe if you work with your spouse but that s relatively rare and then we only have one false positive So it s hard to be in that bed with more that two people don t know Laughter Maybe you have like don t know mean now all of them are going to have phones maybe if there are artists don t know but laughter so we re now in a phase of social science research where we ve got the data or at least the government has the data At least the phone company has the data At least these guys have the data Ask them for graphs The issue is how will we manage all this data And it s true that these are this is data about our connections and our connections come in flavors and we know that for example that there are people who have strong ties and Barry Wellman has really done an enormous amount of work identifying just how many strong ties do you have and did the internet change this numbe
66. user interface I m talking to the camera In user interface design you should be thinking about what your user is most likely to do and that should be one of the simplest things that you can do with that tool Most users are going to eventually say can change the size and color of the node How would do that manually Well you probably want to do it in the autofill thing but what we ve done is made it really easy to change attributes of these nodes All you have to do is find the row for let s say Nancy there s Nancy s row and the current shape for Nancy is image and the size is ten and the opacity is 90 and that s the image file And she has a label and it s her name and there s a tooltip and there s this little description there and if we wanted to change the color that would be easy would just type something like red no not there would type it there red enter If now refresh the graph her node is going to turn red I ll refresh the graph So that s supposed to be easy and we ve tried to make it as easy as possible And so you can now get data from a variety of sources Maybe data that you already have that you can import or other data that s up on the web in various social media repositories you can get that data You can analyze that data In the automate feature we will stack up in the right order all the things that you need to do to get your graph and run it through a process that takes it from a raw state to cooked And that
67. will show you the connections among them or if you deliver a list up to 10 000 names we will give you the network of connections among them But the one that use was the one that use was a map of SSW or SSW 2012 from Twitter from the search network So what we do is we throw a query any query that you can use in a Twitter search box so that means ands ors nots and minus sings quotes all that kind of stuff works We throw that at Twitter search Twitter throws back up to 1 500 tweets Those tweets have authors We iterate over the author names We ask Twitter about each author We say Who follows that author and who does that author follow And then we ask NodeXL to run through that list and say Of all the people who actually tweeted about X who follows each other And what else could we say Well we also click the Replies to mentions and tweet edges A reply is when I say your name at the beginning of a tweet at your name blah blah blah That s a reply A mention is blah blah blah at your name That s a mention of you and we differentiate those edges So we pull in those edges We also say Give me the data that Twitter has and the information about URLs and we ve just added a feature that does what we refer to as the double unwrap So when you now put a URL into Twitter Twitter replaces that with letter referred to as the tco URLs T dot co slash some hash That s what all URLs in Twitter now look like But if you resolve the t
68. writing a user manual for these bold features is so hard and so we gt gt So there is another volume coming gt gt Yeah yeah gt gt That s what I m going to do with my summer vacation gt gt Yeah gt gt Well you know writing the book was a major ingredient of the success of the tool And it s just if you don t have documentation your tool is of no use to almost anybody So it s the castor oil of software development right You know writing the documentation nobody really wants to take it but it s supposed to be good for you So just to illustrate this is the before and after so this is hey let s make the big splatter graph This is what NodeXL will do and almost all network tools will do if you just hand it a graph This is what happens after the application of Group In a Box And we believe that the combination of several of these techniques like other little techniques like many network analysis tools visualization tools when you feed them isolated vertices we ll do almost exactly the wrong thing with them So inaudible the layout algorithm for networks will create what call the asteroid belt which all of these isolates get pushed away from everything else and so they go to the edge of the screen They kind of form an oval if you have a rectangular screen Haroll Corin phonetic says all isolates share the same structural location Therefore will plot them right on top of each other We say to these algorithm
69. x3md that would be Alan Neustadl had a mention of Trent M Kays and that came in the form of a re tweet of Trent And you were saying that you re coming to our Webshop and Alan re twitted you on the 17th of August at 10 18 p m universal time code So GMT zero We have all of the URLs that s the URL that points to the tweet and if there were URLs in the tweet you ll see that we populate the separate column with the URLs in the tweet We also do that for the hashtags We also give you the longitude and latitude of the tweet if it s present In here we see that Viege phonetic who s Viege Are you here Maybe somebody else And have a feeling based on those lat longs that s not somebody here were definitely not at negative one So we could take this and walk it over It s where All right Okay so let s go to maps and it s kind of interesting thing that you can take any of these tweets and you just type the lat long in and you hit this And we should then promptly zoom directly to this guy s inaudible Laughter And I ll tell you it s kind of frightening because where watch where it s going to put the pin When look at the tweets that have done using this it can tell the difference between the tweets like Twitter from the kitchen the bedroom and the bathroom Laughter don t want you to know which ones are which So Laughter so yeah when you zoom down you end up really really seeing It s not just in the street or in front of t
70. y between the two lots of polarization So would say that left and right have both very aggressively moved into Twitter Now does it matter and this is to your point does it matter Is anybody changing anybody s mind in this place doubt it got to say part of the downside of the research is that I m not reading the content from the part of the world that usually you know don t want to hear it don t want I don t watch Fox News don t read National Review don t want to know what they know But now am I m reading all these stuff And it has had no effect on my political views if anything Well I ll say it has entrenched my political views gt gt might say you should ask Dr Jen Golbeck who did study two three years ago she read every tweet written by a member of Congress Laughter Some 6 000 gt gt All three of them gt gt And there was an overwhelming strength on the Republican side This Republicans had organized early on to push the members of Congress to tweet And so there was a strong you know strong and heavy dominance on the Republican side think that s balanced out and it d be interesting what Jen Golbeck has to say about that now I m not sure we can say anything else There was early on the TCOT top as I learned from Mark the top conservatives on Twitter but then that was T lot top liberals on Twitter So you can look for those groups and see the sizes of that would say from what I ve see
71. you in a second but that s a deep if that s true I m deeply concerned I m more concerned than merely methodologically flawed if you re saying that they re actually manipulating at that I m not sure if that s a substantiated allegation but that would be very serious gt gt Correct and don t think that though you can necessarily inaudible approve but that might be true But it does it is at least some analysis I m sure they can work out since they are kind of promoting this as the analysis gt gt A couple of the caveats and l Il be right there with you think it s eight to ten percent of US population are Twitter users Skews in all sorts of SES directions towards you know why yes do have a Honda Civic What do you know And yes am thinking about a Prius do have an iPhone You know do have a Bachelor s and yes do have a graduate degree so you know there is a big skew but my argument is this If a bunch of GOP AOK people say Today we re going to tweet a lot of negative about Mr Obama Or if a bunch of GOP not so much people say Today is the day we re going after Romney And that moves these numbers does that mean anything If the people who are against you are against you then that s not news If the people who are against you got noisier that s not even that much news but that s being reported as an actual change in the aggregate and will argue that at the last few hours and days before the election there i

transcript - University of Maryland at College Park

Contents

Download Pdf Manuals

Related Search

Related Contents