User Guide - SAP Service Marketplace
Contents
1. 5amples K5C H Browse Events fie view csv P d a 2l Browse Gn Wr 4 Click the Next button Describing Events Data The screen Events Data Description lets you describe your Transaction data offering you the same options as the screen Data Description For Sequence Coding to function properly there must be a variable in the Transaction data set that is the same as the primary key declared for the Reference data set referred to as a Join Column The name of the variable can be different but the storage and value must be the same The values of this variable need not be unique since each Reference key can have O 1 or several associated transactions In addition to a suitable join column the Transaction data set must have at least one datetime variable The datetime variable will be used by Sequence Coding to order the transactions One of the datetime variables must absolutely be ordered and declared as such by setting to 1 the Order column for this variable in the description file When the data source comes from a database Infinitelnsight uses a query with an order by on the variable set as Order to retrieve the data But when the data source is a file txt csv Infinitelnsight verifies if the variable set as Order is actually ordered in the file if not an error message is displayed For detailed procedures on how to set parameters on this screen see Describing the Data on page 92. Inthe advanced parameters keep 75 of the hits Note To know how to set the parameters go to section To Set the Parameters on page 16 in scenario 1 Selecting Sequence Coding Statistics For this Scenario In this scenario you decide to calculate for each session which pages have been visited on the web site and what page led the internaut to another By adding page transactions count to the model more information on the internauts behavior will appear You decide to calculate for each session which pages have been visited first and last on the web site and what pages had been visited in between That way you should be able to determine when a visitor is going to leave your web site and decide on which pages to make a 5 reduction offer to keep the visitor and encourage him to make a purchase 33 Scenario 2 Predict End of Session Using Intermediate Sequences otep 3 Generating and Validating the Model You must use the following settings For the variable Page select the function FirstLast which will create two states columns for each session one containing the first page visited the other the last page visited Note To know more about Sequence Coding Statistics go to section Selecting nfinitelnsight Explorer Sequence Coding Statistics see Selecting Sequence Coding Statistics on page 22 in scenario 1 Checking the Transactions For this Scenario After the transactions are checked Sequence Coding should ha
3. CUSTOMER End User Documentation Document Version 1 0 2014 11 S AN ua SS Lg ANS y F ESA ws y L E ava f j 1 j i M Ia SU ee Wear ii N UL i N Ya NN a o A Y H d N K N ni adi s M t x N Ne N LN v A A r f i p A P TA UE i Nc vw M 1 N E N Table of Contents Introduction to Application Scenarios 3 Oh T 4 hic277 1 4 XX QRm 4 Introduction to Sample Files 5 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts 6 TIN EE TED T RE pelectmo a Data SOUFOOu ebrietate tnni taper bui dicia iuto s Mee abri oan cri eM UM a 8 ASS CMOS WS P IC E Y 9 oe ct ESVEIES JU Ed oiiorosdden unie rei E sc eu de bur 13 IBI re iori ass cung Boc C Y 14 Siep 2 Delinme the Modeling Pardel eT S ceis ran ERI A TUE CEREASIS OVE E NUAVERU WAV EVU RUE 15 Seting Sequence Coding Paramieletrs 2 e omen ete t re ae Ep quee oU Eg E HI URINE FEE UN EDU E UP ERAS ed 15 5elec ng Sequence Coding S Latis IQ S dures eedote omi ute emi t HE aE 22 Checking the TranSaCu OU cicecscassndevescinsescecsnncvencusadivedessesionadeddacnedawctnadsndddsadondesaacheddicusdesetencunndics 24 Selce Variables sree i E a a EEE O TNE
4. Each variable is described by the fields detailed in the following table The Field Gives information on Name the variable name which cannot be modified Storage the type of values stored in this variable Number the variable contains only computable numbers be careful a telephone number or an account number should not be considered numbers String the variable contains character strings Datetime the variable contains date and time stamps Date the variable contains dates Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 1 Selecting the Data The Field Value Key Order Missing Group Description Structure Gives information on the value type of the variable Continuous a numeric variable from which mean variance etc can be computed Nominal categorical variable which is the only possible value for a string Ordinal discrete numeric variable where the relative order is important Textual textual variable containing phrases sentences or complete texts Warning When creating a text coding model if there is not at least one textual variable you will not be able to go to the next panel whether this variable is the key variable or identifier for the record 0 the variable is not an identifier 1 primary identifier 2 Secondary identifier whether this variable represents a natural order 0 the variable does not represent a n
5. checkout process The presence of order5 tmpl indicates that a purchase has occurred Since the goal of the analysis is to gain new insights into what behaviors lead to a purchase these order pages and other similar information must be excluded from the analysis Vl To Select a Target Variable On the screen Selecting Variables in the section Explanatory Variables Selected left hand side select the variables you want to use as target variables KXEN Modelling Assistant New Model with Sequence Analysis E i E E EN ini xj Selecting Variables Target Variables _ Alphabetic Sort contact contactHowToOrder html Page cc CU contact contactMain html Page an Wr 26 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters MV To Exclude Explanatory Variables 1 Onthe screen Selecting Variables click the button Open a Saved List located under the section Excluded Variables i KXEN Modelling Assistant New Model with Sequence Analysis Selecting Variables xc CO Member creditinfo html Page c CO Member frameMember html Page c CO Member frameMemberHome html Page xc CO Member frameSecure html Page xc CO Member navigationMember html Page c CO Member arderstat html Page xc CO Member arderStatResults html Page kc CO Member signIn html Page xc CO Member signUp html Page xc CO Member update html Page c CU contact contactFaq html Page kc CO contact
6. 26 Setting the Number of C Iustets uoo cioe tren oe Preterea e as too isin kaniinia iiaeaa vn aia aai ia 27 Step 3 Generating and Validating the Model esses eene eene 28 Gencratina the 28 V alidatim the Model sersseseiisssnssasn aagi a a TE onia 29 Step 4 Analyzing and Understanding the Model sse eee eene 31 SCE DOSOLIDEOBS siessen nonien onae A MEN nea ERNST ae 31 Scenario 2 Predict End of Session Using Intermediate Sequences 32 Step d Selecting the EE 33 Step 2 Defining the Modeling Parameters eese eene eene near nnne ness n nns 33 Setting Sequence Coding Parameters retient rose edP FEE E pEES UE dUr dT ERE PHI UPI E EUR LE ePEESQdEdS 33 Selecting Sequence Coding Statistics odes e ero derum een retenta abicadevasidelavsecademnntedolessetsdaeass 33 Checking the Transactions s ssseeeoeeeeeeessssssssssssseceteerrrsssssssssssseettreeesossssssssssseeetrerreessssssssseees 34 Soken INCISO LEER 34 Step 3 Generating and Validating the Model esses eene eene enne 34 Gencrdune he Model PR TTRRRR m 34 vatidaune he Mode sssr A EEEn EE EE ARER 35 Step 4 Analyzing and Understanding the Model sees 36 Contributions by Variables eeseeessssssssssssssssseseee nennen nennen nennen eee sssnee nn nnn nnn nennen nnns 36 Significance O
7. For this Scenario Use the description file file view desc csv 14 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters M To Describe Events Data 1 2 3 4 On the screen Event Data Description click the button Open Description The following window opens t Load a Description for file_view csv Data Type Folder Description e E Browse cx Cancel In the window Load a Description select the file file view desc csv Click the OK button The window Load a Description closes and the description is displayed on the screen Event Data Description i KXEN Modelling Assistant New Model with Sequence Analysis Description file view desc csv De ee ae ae Te oe men one oem Sem sting mw Q8 L iuune e ee R O O o 3mm feme ws p 1 d4Pue pem ps p pb f 9 Add Filter in Data Set e Analyze e Open Description LJ Save Description Q Ten Baa Ee vibe Note that the Order column is set at 1 for the Time variable thus indicating that this variable is used as a natural order Click the Next button Step 2 Defining the Modeling Parameters Setting Sequence Coding Parameters The screen Sequence Analysis Parameters Settings enables you to set some Sequence Coding parameters by performing the following tasks 19 Scenario 1 Segment Visitors
8. Note The folder selected by default is the same as the one you selected on the screen Data to be Modeled 4 Inthe Description field select the file containing the data set description with the Browse button Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 1 Selecting the Data 5 Click the OK button The window Load a Description closes and the description is displayed on the screen Data Description KXEN Modelling Assistant New Model with Sequence Analysis fe Description session_purchase_desc csv _ Add Filter in Data Set uj Analyze e Open Description LJ Save Description l A View Data an xr UA 6 Click the Next button Selecting Events Data The screen Events Data lets you specify the data source to be used as the Transaction data set For this Scenario The Folder field should already be filled in with the name of the data source that you specified on the Data to be Modeled screen Selectthe file file view csv Vv To Select Events Data 1 Select the format of your data source Text Files ODBC 2 Inthe Folder field specify the folder where your data source is stored 13 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 1 Selecting the Data 3 Inthe Events field specify the name of your data source i KXEN Modelling Assistant New Model with Sequence Analysis Events Data Source Data Type rext Files Folder
9. data set For this Scenario The model generated possesses 29 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 3 Generating and Validating the Model A quality indicator KI equal to 0 98 Arobustness indicator KR equal to 0 99 KXEN Modelling Assistant Purchase session purchase Training the Model Overview Model Purchase session purchase Initial Number of Variables 2 Number of Selected Variables 110 Humber of Records 50581 Building Date 2012 05 15 14 57 30 Learning Time 1mn 44s Engine Mame Kxen SmartSeqmenter Author denise o Minimum Requested Number of Clusters 10 Maximum Requested Number of Clusters 10 SQL Expressions enabled Suspicious Variables Detected Yes Data Aggregation Kxen SequenceCoder Events Data Total Number of References Humber of Matching References Humber of Processed Transactions Total Number of Transactions Nominal Targets Target Key 1 0 Frequency 95 95 1 Frequency 4 05 Performance Indicators Target Purchase kc Purchase Cluster Counts Initial Number of Clusters 10 Final Number of Clusters 10 This means that Clustering found a reliable grouping KR is greater than 0 90 that does a reasonable job of partitioning the purchasing visitors and the non purchasing visitors KI of 0 98 It is safe to look at the descriptive results of the segmentation to gain insight Scenario 1 Segment Visitors to Understand Purchas
10. list of session IDs and whether each session has led to a purchase or not This will be referred to as the Reference data set for Sequence Coding A Sequence Coding Reference data set must have a single variable unique primary key If the primary key is non unique or spread out over several variables Sequence Coding will not function properly Vl To Select a Data Source 1 Onthe screen Data to be Modeled select the data source format to be used Text files ODBC p KXEN InfiniteInsight Data to be Modeled Default Mode Data Set Factory Mode Data Type Text Files Folder Samples ql Browse baase j O aL E Browse di Cutting Strategy create a Target Metadata No Single Metadata Repository Enabled um 2 Use the Browse button on the right of the Folder field to select the folder where you have saved the sample files 3 Click the Browse button next to the Estimation field and select the file session purchase csv The name of the file will appear in the Estimation field 4 Click the Next button Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 1 Selecting the Data Describing the Data Why Describe the Data Selected In order for the Infinitelnsight features to interpret and analyze your data the data must be described To put it another way the description file must specify the nature of each variable determining their Storage f
11. model has been generated you must verify its validity by examining the performance indicators The quality indicator KI allows you to evaluate the explanatory power of the model that is its capacity to explain the target variable when applied to the training data set A perfect model would possess a KI equal to 1 and a completely random model would possess a KI equal to O The robustness indicator KR defines the degree of robustness of the model that is its capacity to achieve the same explanatory power when applied to a new data set In other words the degree of robustness corresponds to the predictive power of the model applied to an application data set For this Scenario The model generated possesses 35 Scenario 2 Predict End of Session Using Intermediate Sequences Step 4 Analyzing and Understanding the Model A quality indicator KI equal to 0 70 Arobustness indicator KR equal to 0 98 A KXEN InfiniteInsight ksc_Session_continue_session_purchase ally Model Overview sla Report Type Model Overview Model ksc_Session_continue_session_purchase Data Set session purchase csv Initial Number of Variables 2 Humber of Selected Variables Number of Records Building Date Learning Time Engine Name 2012 07 09 14 41 35 4mn 10s Kxen RobustRegression Author denise o Modeling Warnings Monotonic Variables Detected Yes Data Aggregation Kxen SequenceCoder Events Data file view csv Tot
12. value is between the two selected dates will be used Between two date columns Only the events for which the Log Date Column value is between the values of the two selected date columns will be used For example you can select the date columns corresponding to the beginning and the end of a trial period dates that can be different for each customer Relative to a date column Only the events for which the Log Date Column value fits in the range defined with respect to the selected date column will be used For example you can use the purchase date of a credit card as the reference and select all events that occurred in the three months leading to this date WARNING Be careful when choosing a period the selected period must contain events existing in the data set or else you will obtain aberrant results for your model negative KI KR equal to 1 vl To Use All the Events Keep the Infinite option Time Window Infinite qo oo SS xo SOSA C Between two dots coms Fn To C Relative ma dotecolmn Date semeen E e M To Use Only the Events Occurring in a Fixed Time Window 1 Check the Fixed option Time Window Infinite Fixed From 2012 07 09 15 18 44 gi To 2012 07 09 15 18 44 aii C Between two date columns From z To 7 C Relative to a date column Date z Between 1 and Month z Advanced 2 Inthe From field select the date before which no events should be used 3 Inth
13. wjer tis id e M RR RR 4 sje U o P PERENNE E E I E E E E T EAT ANAA E OEE E E E T E E E E EEN 4 Introduction to Application Scenarios Scenario 1 Scenario 1 You will start by using Sequence Coding to create counts of each Web page that was viewed by each visitor followed by a targeted segmentation with purchase as the target This will give you a simple description of the different groups browsing your Web site and the different conversion rates for each group Scenario 2 In this scenario you want to predict when a visitor is goingto leave your Web site Your idea is to offer a 5 coupon to visitors who are likely to leave in the hope of increasing the site stickiness To achieve that you will create a Sequence Coding model using intermediates sequences with the FirstLast option for the pages viewed The intermediate sequence option will automatically create an appropriate target variable for determining which behaviors indicate the end of a session Introduction to Sample Files Scenario 2 Introduction to Sample Files This data set contains a single day of Web traffic from an E commerce site in December 1999 The site content was served by a Broadvision server but no cookies or login was required making the sessions effectively anonymous File session purchase csv session purchase desc csv file view csv file view d
14. F Ate COMES ioooido dri iioi ted FIR NERIS CEN OI iode RE SUR Imam SPEM P in a COSE ar EEn aai 37 Introduction to Application Scenarios Scenario 1 Introduction to Application Scenarios In these scenarios you are the Marketing Director of an E Commerce company and you want to increase the profitability of your Web site You have the budget to launch a major marketing initiative but you re not sure what kind of campaign would be the most effective Due to market pressures you only have the time and money to test a few campaigns before launching a major initiative The two key metrics that are being used to measure the performance of the Web site are the conversion rate and stickiness The conversion rate of a site is the percentage of visits that result in a purchase At this time your Web site has a conversion rate of 496 meaning that 4 out of every 100 visitors purchase at least one item The stickiness of a Web site is a measure of the number of pages viewed by each visitor The more pages a visitor views the more likely they are to purchase something Your Web site is averaging about 10 pages per visit In order to achieve rapid insight into the different groups of visitors to your Web site you have decided to use nfinitelnsight Modeler Segmentation Clustering to group the population with respect to their buying behavior and site abandonment The goal of the analysis is to get descriptions of the groups of visitors who tend to purch
15. al Number of References Number of Matching References Humber l eProcessed Transactions Total Number of Transactions Nominal Targets 50561 50537 532498 532498 Target Key O0 D Frequency 9 58 1 Frequency 90 42 Performance Indicators Target ksc Session continue rr ksc Session continue Kl KR Gr a Previous This means that Classification Regression found a robust model KR is greater than 0 90 that does a reasonable job of predicting the end of a session KI of 0 70 It is safe to look at the variables contributions to gain insight Step 4 Analyzing and Understanding the Model Contributions by Variables The following graph presents the variables contributions Scenario 2 Predict End of Session Using Intermediate Sequences otep 4 Analyzing and Understanding the Model KXEN InfiniteInsight ksc Session continue session purchase ally Contributions by Variables Asada soe Chart Type Maximum Smart Variable Contributions MaxContrib b L1 Ae ge ge ge Eo a ge ge ge ge v ge ge d oue E us P a ex eat Qe 9 av gat tal que eat ew oe oie eat eX eat M LAM M uM MEM uu uM tthe athe ath gah gat fd JD GE Lut f x A MP VUE a Saa a ger SU gU AT s Variables The pages having the more impact positive or negative on the buying act are listed in the following table Page viewed This variable indicates KSC Page LastState the last page the internaut has viewed befo
16. ant to calculate on transaction or event data For this Scenario You decide to calculate for each session which pages have been visited on the web site That way you should be able to determine and understand which pages led the visitors to make a purchase 22 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters You must use the following settings For the variable Page select the function Count which will create a state column for each page visited vl To Select Sequence Coding Statistics 1 TheSequence Analysis Variables Selection for Functions screen lists all the variables for which statistics can be calculated For each variable listed select the functions to use You can choose among the three functions Count CountTransition and FirstLast i KXEN Modelling Assistant New Model with Sequence Analysis Gn Wr 2 Click the Next button Operations Definition Several standard Sequence Coding columns are created for each reference ID For reference Ids that have no transactions associated with them the standard Sequence Coding columns will have null values KSC Start Date The timestamp of the first transaction in the log for each reference ID KSC End Date The timestamp of the last transaction in the log for each reference ID KSC TotalTime The seconds between the KSC Start Date and KSC End Date KSC Number Events The number of transactions in t
17. ase items frequently and the indicators that a session is about to end You already know the following basic facts about your Web site An average of 50 000 visitors come to the Web site each day For the 2000 sessions that result in a purchase each day the average amount spent is 181 The average profit margin for the Web site is 596 so each purchase results in an average profit of 9 05 resulting in 18 100 of profit per day There are four main entry points for the site The home page the members home page the sweepstakes page and the specials page The checkout process has five steps all with the word order in the file name Your site does not use cookies or require a login for your members so each session is effectively anonymous unless a purchase is made The information that is available for analysis consists of the Web logs Your DBA has pulled out a list of the sessions from a single day of traffic along with a flag indicating if the session resulted in a purchase the existence of order5 tmpl in a session indicates a purchase Along with the list of sessions the parsed log from the day is also available Since the information from the Web log is not aggregated for analysis you will need to use the nfinitelnsight Explorer Sequence Coding prior to running the nfiniteInsight Modeler Segmentation Clustering or Infinitelnsight Modeler Regression Classification IN THIS CHAPTER
18. at will indicate the end of the time window In the last drop down list enter the unit to be used to define the time window For example if you have set the parameters Date CardPurchaseDate Between 3 ando Month only events occurring in the three months leading to the date of purchase will be kept for each customer 19 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters Understanding Advanced Parameters The advanced parameters allows you to configure the following elements the prefix to be added to Sequence Coding generated variables the location where the temporary files generated by the modeling are stored the amount of information that will be kept for the modeling Sequence Coding Generated Variable Prefix You can define a specific prefix that will be used to identify variables created by Infinitelnsight Explorer By default this prefix is set to ksc i KXEN Modelling Assistant New Model with Sequence Analysis Advanced Sequence Analysis Parameters Sequence Analysis Generated Variable Prefix Prefix ksc Filter Transitions greater than z Previous Storage Type When creating a model Sequence Coding generates large quantities of temporary columns you can select whether the data generated will be stored in a memory space or on a disk The option In memory is selected by default 20 Scenario 1 Segme
19. atural order 1 the variable represents a natural order If the value is set at 1 the variable is used in SQL expressions in an order by condition There must be at least one variable set as Order in the Event data source Waming If the data source is a file and the variable stated as a natural order is not actually ordered an error message will be displayed before model checking or model generation the string used in the data description file to represent missing values e g 999 or Empty without the quotes the name of the group to which the variable belongs Variables of a same group convey a same information and thus are not crossed when the model has an order of complexity over 1 This parameter will be usable in future version an additional description label for the variable this option allows you to define your own variable structure which means to define the variables categories grouping Viewing the Data To help you validate the description when using the Analyze option you can display the first hundred lines of your data set 10 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 1 Selecting the Data M To View the Data 1 Click the button View Data A new window opens displaying the data set top lines DataSet O Data statistics oes mot eaen NETTEN rerasane oea mee 2st Sie ju a Eee amah 7 a5 Private 160187 sare 2 u
20. contactHowToOrder html Page kc CO contact contactMain html Page A bel C Alphabetic Sort gj Open a Saved List jp um 2 The window Load Excluded Variables List opens In the Variables field select the file containing the variables to skip X Load Variables List E X Data Type Text Files Folder 5amples K5C H Browse File session _purchase_skip csv d H Browse OK Cancel 3 Click the OK button the window closes The list of excluded variables has been populated Setting the Number of Clusters Before generating the model you need to set the number of clusters you want to create For this Scenario Set the number of clusters to 10 which is the default number Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 3 Generating and Validating the Model M To Set the Numer of Clusters In the panel Summary of Modelling Parameters type the number of clusters you want to generate in the field Find the best number of clusters in this range i KXEN Modelling Assistant Purchase session purchase Summary of Modelling Parameters Previous Generate Step 3 Generating and Validating the Model Generating the Model Once the modeling parameters are defined you can generate the model Then you must validate its performance using the quality indicator KI and the robustness indicator KR fthe model is sufficiently powerful you can analyze the res
21. e Behavior Using File CountsStep 4 Analyzing and Understanding the Model Step 4 Analyzing and Understanding the Model Segment Descriptions On the screen Cross Statistics you can look at the logical definition and or the cross statistics of each variable to gain an understanding of what kind of visitors belong to each cluster Three clusters are particularly informative for your business problem which is to determine which kind of population you should try to attract to increase your profit the two clusters that have the highest conversion rates the cluster that has the lowest conversion rate The chart below summarizes these clusters and gives them each a label based on the cluster definition Freq eme Demon fiaa 1 9 31 4 shop shipChart html 0 5 Shippers 3 5 25 96 welcome html 1 20 11 8 0 1 Iholiday holidaySweeps tmpl 1 The cluster Shippers is defined by sessions in which the shipping chart shop shipChart htm has been seen between 1 and 5 times Actually this cluster does not give you much information It just tells you that visitors that go to the shipping chart will probably make a purchase which is rather logical If you don t intend to buy why would you look at the shipping information The cluster Members is more informative It shows that people visiting the member home page welcome html are more likely to buy This is an interesting piece of information It means that members are m
22. e To field select the date after which no events should be used Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters M To Use Only the Events Occurring Between Two Date Columns 1 ooh 2 Check the option Between two date columns Time Window Infinite C reed von A 3j o aS Between two date columns From To Shame sie Oo ME In the From field select the date column containing the date before which no events should be used In the To field select the date column containing the date after which no events should be used To Use Only the Events Occurring in a Range Relative to a Date Column Check the option Relative to a date column Time Window C Infinite Fed von A vo A C Between two date columns From To Relative to adate coume Date E Between 0 ana 18 Month 7 In the Date list select the column that contains the date to use as a reference for the time window In the Between field enter the number of units that will indicate the start of the time window The following table sums up the values you can use to define the beginning of the time window Value Significance negative integer the time window begins before the reference date 0 the time window begins at the reference date positive integer the time window begins after the reference date In the and field enter the number of units th
23. ed Variable Prefix me 1 ee for Holding Counting Information Filtering Percentage of Hits Kept 75 Understanding Infinitelnsight Explorer Sequence Coding Parameters J oining Your Data To aggregate the reference data with the events data you have to join both tables and indicate which column of each table corresponds to the reference ID In the fields Columns for Join select the variables corresponding to the customer ID in both data sets The information contained in both selected variables must be the same In the field Log Date Column select the variable corresponding to the date and or time of the log data Calculating the Intermediate Sequences The mode Intermediate Sequences provides you with additional information about the transitions and sequences existing in your data sets order of the steps details of the steps continuity of the session for each step 17 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters Filtering the Events by Time Window The section Time Window allows you to filter the events on which the model will be built by setting a period defined either by fixed dates or by values existing in the data set The following options are available to filter the events data set Option Description Infinite No time window is defined all the events will be used Fixed Only the events for which the Log Date Column
24. esc csv session purchase skip csv session continue skip csv Description list of sessions and binary purchase target 50581 rows description for session purchase csv log of files requested from Broadvision server 532860 rows description for file view csv variable skip list for Scenario 1 these are the variables where the value would not be known until the session had ended variable skip list for Scenario 2 These sample files can be downloaded on nfinitelnsight Sample Files Download Center http www kxen com sample_data Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts ocenario 2 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts 1 In Infinitelnsight main menu select the option Perform a Sequence Analysis in the Explorer section KXEN InfiniteInsight X InfiniteInsight Version 6 1 0 Explorer Modeler Create or Edit Explorer Obj Create a Classification R Create a Clustering Model Create a Time Series Analysis Perform a Sequence Analysis Load a Model Perform a Text Analysts um ence Analysis Aggregate Events into a Series of Transitions Create a Social Network Analysi Open the Data Viewer Load a Social Network Analysis Model Perform a Data Transfer List Distinct Values in a Data Set Get Descriptive Statistics for a Data Set 2 The screen Add a Modeling Feature is displayed KXEN InfiniteInsight Add a Modeli
25. he log associated with each reference ID In addition to the standard nfiniteInsight Explorer Sequence Coding columns three types of operations are available Count Count the transitions First and last 23 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters Count When you select the Count option Sequence Coding creates a new column for each value of the inserted variables Count encodes the sequences using one column per valid category in the specified nominal column Each valid category is referred to as a state Categories that are seen only once for the transactions associated with the reference id present in the Estimation data set are discarded CountTransition When you select the CountTransition option Sequence Coding creates a new column for each transition of categories in the selected data set CountTransition encodes the sequences using one column per valid pair wise category transition in the specified column Each valid category transition is referred to as a state transition State transitions that are seen only once for the transactions associated with the reference id present in the Estimation data set are discarded A separate KxOther column will be created for rare transitions using the threshold set by the Filter slider bar in the same way a KxOther column is created for the counts FirstLast The FirstLas
26. ly analyze the sequences or add extra transformations such as a Classification Regression J nfinitelnsight Modeler Hegression Classification or a Clustering Segmentation nfinitelnsight Modeler Segmentation Clustering IN THIS CHAPTER Step 1 Selecting BiU ADI RETE R Step 2 Defining the Modeling Parameters cccccecccccccesseceecceesecceeceeseceeeceuseceecseeseceeesaeseceesseeeeceeesaueceeeseaneeees Step 3 Generating and Validating the Model cccccesseccceceeseceeeceeseceeeceeseeceeseesceseseeaceesseeaeceesseaeceeeseaeees Step 4 Analyzing and Understanding the Model sseeeessssssssssssseeeeneeeenne nennen nennen nnns nna nnns Step 1 Selecting the Data To know how to select and describe the data go to section Selecting the Data on page 8 and Describing the Data on page 9 in Scenario 1 For this Scenario Select the Random cutting strategy Use the file session purchase csv as the reference file and use the file session purchase desc csv as its description file Selectthe file file view csv and use the description file file view desc csv Step 2 Defining the Modeling Parameters Setting Sequence Coding Parameters For this Scenario Select the SessionID column as the join column for both the log and reference data sets Select Time as the Log Date Column Check the option Intermediate Sequences
27. m contactsap 2014 SAP SE or an SAP affiliate company All rights reserved No part of this publication may be reproduced or transmitted in any form or for any purpose without the express permission of SAP SE or an SAP affiliate company The information contained herein may be changed without prior notice Some software products marketed by SAP SE and its distributors contain proprietary software components of other software vendors National product specifications may vary These materials are provided by SAP SE or an SAP affiliate company for informational purposes only without representation or warranty of any kind and SAP or its affiliated companies shall not be liable for errors or omissions with respect to the materials The only warranties for SAP or SAP affiliate company products and services are those that are set forth in the express warranty statements accompanying such products and services if any Nothing herein should be construed as constituting an additional warranty SAP and other SAP products and services mentioned herein as well as their respective logos are trademarks or registered trademarks of SAP SE or an SAP affiliate company in Germany and other countries All other product and service names mentioned are the trademarks of their respective companies Please see htt W Sap C X additional trademark information and notices i I i ve
28. ng Feature Add a Classification Regression Standalone Data Transformation gai Cancel Previous Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts ocenario 2 3 Click on the option Add a Clustering Note When building a model you can either simply analyze the sequences or add extra transformations such as a Classification Regression J nfinitelnsight Modeler Hegression Classification or a Clustering Segmentation nfinitelnsight Modeler Segmentation Clustering IN THIS CHAPTER otep Ee e 10 0 RR RP tc eee ee enn ee en 8 Step 2 Defining the Modeling Parameters ccccccccssseeeececeeeseeeeeceeceeeeeeeseeeeeeessseaeeeeeeeeeeseaeseeeeeeessseaeeeeeeeessaas 15 Step 3 Generating and Validating the Model cccccssssccceceeecceeceeeecceeceeseeceeseeseceeeeeeseceesseeseceeesaueceesseaeees 28 Step 4 Analyzing and Understanding the MOdel cccccccccessecceeceeseeeeeceeeeeecceeseceessueeeceesseeeceeesaugeceeeeeaneeees 31 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 1 Selecting the Data Step 1 Selecting the Data IN THIS CHAPTER SCICCING a Dala SO UNCC ERE TER E 8 Bi teglelteBi 28 i MNRMR 9 FECE eN Dc RE Omer 13 BS SGC Eyents DII ENTRE TERR 14 Selecting a Data Source For this Scenario The file session purchase csv contains a
29. nt Visitors to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters Xx KXEN Modelling Assistant New Model with Sequence Analysis 1 oe Analysis Generated Variable Prefix mee o o i E Storage Used for Holding Counting Information In Memory On Disk Ao Cancel Previous Filtering the Events The Filtering option allows you to group rare categories into a single category labeled KxOther It is very common for transaction logs to have many infrequently occurring categories that by themselves will not make reliable predictors A predictive benefit can often be achieved by combining these rare categories into a single group The Filtering slide allows you to select the categories to keep as separate columns based on percentage of the overall transaction log The categories corresponding to the remaining percentage of transactions are grouped in the KxOther column which is automatically generated by Infinitelnsight Explorer Sequence Coding Advanced Sequence Analysis Parameters Sequence Analysis Generated Variable Prefix wea Storage Used for Holding Counting Information v InMemory 6 On Disk Filtering Percentage of Hits Kept 905 D 10 20 30 40 50 60 70 80 30 100 Filter Transitions greater than gai Cancel Previous Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters For example if you se
30. ore likely to make a purchase than other visitors So increasing the number of members should increase your profit The cluster Sweepstakers gives you information on a previous attempt at increasing the number of purchase through a sweepstake You can see that only 0 196 of the people visiting the sweepstake page actually make a purchase You can infer from this that your previous campaign had the effect opposite to the one expected 31 Scenario 2 Predict End of Session Using Intermediate Sequences otep 4 Analyzing and Understanding the Model Scenario 2 Predict End of Session Using Intermediate Sequences 1 Ini nfinitelnsight main menu select the option Perform a Sequence Analysis in the Explorer section KXEN InfiniteInsight X InfiniteInsight Version i 6 1 0 m Explorer In Modeler Create or Edit Explorer Obj Create a Classification Re Create a Clustering Model Create a Time Series Analysis Create Association Rules X Toolkit Open the Data Viewer Perform a Data Transfer List Distinct Values in a Data Set Get Descriptive Statistics for a Data Set 2 The screen Add a Modeling Feature is displayed ec InfiniteInsight Add a Modeling Feature Adda d Regression Standalone Data sformation 3 Click on the option Add a Classification Regression z 32 Scenario 2 Predict End of Session Using Intermediate Sequences Step 1 Selecting the Data Note When building a model you can either simp
31. ormat number number character string string date and time datetime or date date Notes When a variable is declared as date or datetime the KXEN Date Coder feature KDC automatically extracts date information from this variable such as the day of the month the year the quarter and so on Additionnal variables containing this information are created during the model generation and are used as input variables for the model KDC is disabled for Time Series Type continuous nominal ordinal or textual For more information about data description see Types of Variables and Storage Formats in the Introductory Guide to nfiniteInsight How to Describe Selected Variables To describe your data you can Either use an existing description file that is taken from your information system or saved from a previous use of InfiniteInsight features Orcreate a description file using the Analyze option available to you in nfinitelnsight In this case it is important that you validate the description file obtained You can save this file for later re use If you name the description file KxDoc_ lt SourceFileName gt it will be automatically loaded when clicking the Analyze button Important The description file obtained using the Analyze option results from the analysis of the first 100 lines of the initial data file In order to avoid all bias we encourage you to mix up your data set before performing this analysis
32. ponses that it provides in relation to your business issue Otherwise you can modify the modeling parameters in such a way that they are better suited to your data set and your business issue and then generate new more powerful models 28 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 3 Generating and Validating the Model vl To Generate the Model On the screen Summary of Modelling Parameters click the Generate button The screen Training the Model will appear The model is being generated A progress bar will allow you to follow the process KXEN Modelling Assistant Purchase session purchase Training the Model Beginning of learning for Default Building transition vectors Stop Current Task Validating the Model Once the model has been generated you must verify its validity by examining the performance indicators The quality indicator KI allows you to evaluate the explanatory power of the model that is its capacity to explain the target variable when applied to the training data set A perfect model would possess a KI equal to 1 and a completely random model would possess a KI equal to O The robustness indicator KR defines the degree of robustness of the model that is its capacity to achieve the same explanatory power when applied to a new data set In other words the degree of robustness corresponds to the predictive power of the model applied to an application
33. re ending his session KSC Last duration duration of the session from the first page viewed to the previous state KSC LastStepNumber the number of pages the internaut has viewed before ending his session Count holidaySweepsEntry ht the number of time the page holidaySweepsEntry ml access to holiday promotions has been viewed The impact of each page on the purchase is detailed in section Significance of Categories Significance of Categories KSC Page LastState 37 Scenario 2 Predict End of Session Using Intermediate Sequences otep 4 Analyzing and Understanding the Model This is by far the strongest predictor This is similar to a low order Hidden Markov Model where the current state is used to predict the next one Last duration and LastStepNumber The length of the session and the number of pages viewed are also important If the internaut has viewed only one page he has not yet entered the site and may end his session because the site may not seem of interest to him but if he has viewed more than 12 pages he has probably found what he was looking for and will end his session If he has seen between 2 and 11 pages he is probably shopping and thus should continue his session Count holidaySweepsEntry html If the page has been viewed it is a good indicator that the session will continue Since this page is the entry point of a holiday promotion the internaut will at least go to the promotion page 38 WWW Sap co
34. sistant New Model with Sequence Analysis Model Checking Checking transactions file ct ae Stop Current Task 2 When the process is over click the button Show Detailed Log The number of columns created by Sequence Coding is indicated ESC kept 98 state columns for variable Page Date Coder module inserted in the processing chain el Checking is Finished ESC kept 98 state columns for variable Page Date Coder module inserted in the processing chain el Checking is Finished KSC kept 98 state columns for variable Page Date Coder module inserted in the processing chain el Checking is Finished Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters 3 Click the Next button Selecting Variables Once the reference data set the events data set and their descriptions have been entered you must select different variables oneor more Targets Variables possibly a Weight Variable andthe Explanatory Variables For this Scenario Keep Purchase as the target Use the session continue skip csv file to select the variables to exclude This list of variables includes the information that is not known about a session until a purchase has occurred or is very likely to occur For this Web site the checkout process included five order pages The presence of any of the five order pages in the log indicates that they have already started the
35. t option creates two columns the categories of the selected variable from the first and last transactions in the log for each reference ID called FirstState and LastState respectively The FirstState and LastState columns are created automatically when either the Count or CountTransition options are selected Checking the Transactions At this stage nfinitelnsight analyses the data sets and creates a number of new variables or columns Depending on which operations you chose during the previous step Sequence Coding creates four standard columns ksc Start Date ksc End Date ksc TotalTime and ksc Number Events one column for each state if you have selected Count one column for each transition if you have selected CountTransitions Two columns FirstState and FinalState if you have selected Count CountTransitions or FirstLast Six columns LastStepNumber Last date time Last duration Session Continue LastState and NextState if you have selected Intermediate Sequences For this Scenario 24 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters After the transactions are checked Sequence Coding should have kept 99 state columns for the Page variable plus the four standard columns and the FirstState and LastState columns vl To Check the Transactions 1 During the model checking a progress bar is displayed i KXEN Modelling As
36. t the Filtering slider at 9096 it means that the total number of transactions when adding all the categories assigned to separate columns must not exceed 90 of the total number of transactions The categories that make up the remaining 1096 of the transactions will be grouped under KxOther You can also define a threshold so that transitions which duration between two events is higher than the defined threshold will be ignored in the transition count vl To Set a Threshold 1 Check the box Filter Transitions greater than Tu Modelling Assistant New Model with Sequence Analysis Sequence Analysis Generated Variable Prefix Prefixfisc o 2 Inthe number field enter the number of units defining the threshold 3 Inthe drop down list select the unit to be used to define the threshold For this Scenario For the sample data each row of the transaction log represents an HTML file requested by the visitor s browser There are 10184 different files that are requested during the day However by positioning the Filtering slide at 75 only 99 files are retained for separate count columns and the rows with the remaining 10085 files are grouped into the KxOther count This means that the 99 most common files make up 75 of the log and the remaining 10085 files make up only 25 of the log Selecting Sequence Coding Statistics The screen Sequence Analysis Variables Selection for Functions lets you specify the type of Statistics you w
37. to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters Join your reference data with your transaction data Calculate the intermediate sequences Filter your events by period For this Scenario Select the SessionID column as the join column for both the log and reference data sets Select Time as the Log Date Column In the advanced parameters keep 75 of the hits Select Infinite as the Time Window Vl To Set the Parameters 1 Onthe screen Sequence Analysis Parameters Settings select the join column for both the log and reference data sets 2 Select the Log Date Column eta Modelling Assistant New Model with Sequence Analysis Sequence Analysis Parameters Settings Events Data Set Reference Data Source Columns for Join SessionID v SessionID x Events Date Column Time v Intermediate Sequences Time Window Infinite Fixed From 2012 05 15 12 00 55 a To 2012 05 15 12 00 55 yi Between two date columns From J To J Relative to a date column Date 8 Between oa and A Month 3 Click the Advanced button to set the advanced parameters 16 Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 2 Defining the Modeling Parameters 4 inthe Advanced panel slide the filter to 7596 i KXEN Modelling Assistant New Model with Sequence Analysis Advanced Sequence Analysis Parameters Sequence Analysis Generat
38. u E zt LN zc LL B arried civ Exec i5 neme izzape married n c H spee masce iiNeiernered aes 2 Inthe field First Row Index enter the number of the first row you want to display 3 Inthe field Last Row Index enter the number of the last row you want to display 4 Click the Refresh button to see the selected rows A Comment about Database Keys For Sequence Coding to be able to join the Reference and Transaction data sets the Reference data set to be analyzed must contain a single variable that serves as a unique key variable To Specify that a Variable is a Key 1 Inthe Key column click the box corresponding to the row of the key variable Scenario 1 Segment Visitors to Understand Purchase Behavior Using File Counts Step 1 Selecting the Data 2 Type inthe value 1 to define this as a key variable n Description Desc CensusO1 csv Add Filter in Data Set For this Scenario Use the file session purchase desc csv as the description file M To Describe the Data 1 Onthe screen Data Description click the button Open Description The following window Opens KXEN Modelling Assistant New Model with Sequence Analysis Add Filter in Data Set Vg Analyze iJ Open Description LJ save Description l 2 Inthe window Load a Description select the type of your description file 3 Inthe Folder field select the folder where the description file is located with the Browse button
39. ve kept 98 state columns for the Page variable Selecting Variables For this Scenario Use the session continue skip csv file to select the variables to exclude Use KSC Session continue as the target and remove Purchase from the targets Note To know how to select variables go to section Selecting Variables see For this Scenario on page 26 in scenario 1 Step 3 Generating and Validating the Model Generating the Model Once the modeling parameters are defined you can generate the model Then you must validate its performance using the quality indicator KI and the robustness indicator KR fthe model is sufficiently powerful you can analyze the responses that it provides in relation to your business issue Otherwise you can modify the modeling parameters in such a way that they are better suited to your data set and your business issue and then generate new more powerful models 34 Scenario 2 Predict End of Session Using Intermediate Sequences Step 3 Generating and Validating the Model vl To Generate the Model On the screen Summary of Modelling Parameters click the Generate button The screen Training the Model will appear The model is being generated A progress bar will allow you to follow the process KXEN Modelling Assistant Purchase session purchase Training the Model Beginning of learning for Default Building transition vectors Stop Current Task Validating the Model Once the
Download Pdf Manuals
Related Search