Lab 3: Using Analysis Models on Your Site

In this lab you will create a transaction-based Prediction model and use it on your Web site.

This lab contains the following steps:

  • Step 1: Build the Transaction model
  • Step 2: Analyze predictions on your Web site
  • Step 3: Use the Affinity List module

Step 1: Build the Transaction model

  1. Click Start, point to Programs, point to Commerce Server 2002, and then click Commerce Server Manager.

  2. In Commerce Server Manager, expand Commerce Server Manager, expand Global Resources, expand Predictor on <server name>, expand Predictor Service, and then select Model Configurations.

  3. In the Commerce Server Manager content pane, right-click DemoTransactions, and then select Build.

  4. In the Build Model Properties dialog box, do the following:

    Use this To do this
    Name Type Transactions.
    Model type Select Prediction asan analysis model type to build.

    Ee811243.important(en-US,CS.20).gifImportant

    • Use the exact model names for analysis models because ASP pages will use that name when making predictions.
  5. Click Next.

  6. In the Build Model Properties dialog box, in the Measured accuracy sample fraction box, type 0.1, and then click Finish. Wait until the model is successfully built.

  7. In the Commerce Server Manager content pane, right-click Transactions, and then select View Model. Note the following:

    The dependency network for Predictions should look identical to the dependency network for Software Predictions created in Lab 2: Viewing and Interpreting Analysis Models.

Step 2: Analyze predictions on your Web site

  1. Start Internet Explorer, and then go to https://localhost/predictordemo to visit your Predictor Demo site.

    Ee811243.note(en-US,CS.20).gifNotes

    • If the Predictor Demo site fails to load, try doing a site refresh by clicking Refresh from the toolbar.
    • The Predictor Demo site may not function correctly if SQL Authentication was not used during Commerce Server 2002 installation.

    Predictor Demo Site

  2. On the Predictor Demo site, in the Categories list, click Games, click Sports Products, and then click Baseball 2000 in the list.

  3. On the Baseball 2000 product page, click Add to Basket, and then click Basket at the top of the page.

  4. On the Basket page, note the following:

    Cross Sell Recommendations

    • Your product basket includes Baseball 2000.
    • Product recommendations: NFL Fever 2000, NBA Inside Drive 2000, and Office 2000 Standard.

    Ee811243.note(en-US,CS.20).gifNote

    • You can configure the number of recommendations that are displayed on the Basket page by modifying the call to the Predictor function in the ASP code.
  5. Use Commerce Server Manager to go back to the Transactions Dependency Network.

  6. In the Transactions screen, click Find from the toolbar.

  7. In the Node Finder dialog box, select Baseball 2000.QTY from the list, and then click Go to Node. Note the following:

    • If you move the Link Slider (located vertically on the left side of the screen) all the way down and then slowly move the slider up to increase the links strength, the three outgoing links appear from Baseball 2000 to nodes corresponding to the recommended products for Baseball 2000.
    • The recommendations match the order in which the arcs appear.

Step 3: Use the Affinity List module

Using the Affinity List module you can create a list of users and a direct mail campaign to target users who are likely to buy a specific product.

  1. Start the Predictor Demo Business Desk.

  2. In Analysis, click Affinity Lists, and then observe the following:

    • Prediction models are listed on the right-hand side.
    • The Prediction models that appear are either known to predict products or are generated from custom model configurations.

    Ee811243.note(en-US,CS.20).gifNote

    • In order for the export feature to function correctly on a Prediction model generated from a custom model configuration, the predicted property must be named "SKU" with an aggregate column "QTY." If your model does not have these properties, you can manually generate a list using the Predictor API.For more information about generating list manually, see PredictorServiceSiteAdmin::GenerateList.
  3. In the Affinity Lists screen, select Software Predictions from the list, and then click Export on the toolbar.

  4. In the Affinity Lists Properties dialog box, do the following:

    Use this To do this
    List Name Type VJ++.

    This list contains users who are likely to buy Microsoft Visual J++.

    Product Identifier Click the ellipsis [...] button to launch the Product Picker dialog box.

    In the Product Picker dialog box, expand Development Tools, click Visual J++, select Visual J++ Professional Edition, and then click OK.

    Minimal probability for product purchase Type 0.10.

    The Minimal probability for product purchase field is used to specify the threshold criterion for including users in the list.

    The lower the threshold, the greater the number of users that are included. However, users on the list with a lower threshold are less likely (on average) to buy the product.

    The right threshold for your list depends on variables, such as market demand, and specific properties for purchases made on your site. For example, the variety of transactions on your site and the diversity of users who visit your site.

    To generate an effective list, it is recommended that you run a few tests to determine the most effective parameters.

  5. Click OK.

    Ee811243.note(en-US,CS.20).gifNotes

    • The larger the size of the analysis model, the longer it takes to generate a list.
    • When a list is generated for the first time, the Predictor resource caches all the data needed to generate the list in order to make future exports faster.
    • After a list is successfully generated, it appears in the List Manager module in Campaigns.
  6. In Campaigns, click List Manager, and then note the following:

    • VJ++ list appears in the list.
    • You use the List Manager module to create a direct mail campaign to target users who are likely to buy Visual J++. For more information about creating marketing campaigns, see Business Desk Campaigns.

Copyright © 2005 Microsoft Corporation.
All rights reserved.