Using Data Mining

This chapter provides an overview of data mining and discusses how to:

• Configure and use scoring engines, model deployers, and model agents.

• Define models.

• Set up interactive scoring.

• View batch scores.

Understanding Data Mining

PeopleSoft Customer Behavior Modeling integrates data mining—a tool that enables you to analyze the inherent relationships between variables in customer data and generates scores or predictions of customer behavior.

The data mining tool creates predictive models that you can use in marketing, customer interaction, sales force automation, and other applications. For example, you can use customer modeling to find out which customers are overusing call centers; then, you could offer those customers a discount on training. Or you could answer questions such as “Which customers who are on the mailing list are most likely to respond to a marketing campaign?” and “What are the cross-sell and up-sell opportunities?”

The data mining integration feature works via an application programming interface (API) to your chosen data mining tool. (PeopleSoft Customer Behavior Modeling supports Angoss for model creation and deployment.) The API is flexible enough to handle any model size and arbitrary numbers of variables. This is a fault-tolerant, distributed system—all of the activities that the system performs interact asynchronously using multiple scoring engine servers slaved to a model deployer that operates as part of a web server. Clients communicate with the model deployer via PeopleCode API or via XML provided by non-PeopleSoft applications.

Note. Data mining models are created entirely within the vendor application. Once created, the models are deployed using PeopleSoft Customer Behavior Modeling. You use a PeopleSoft Customer Behavior Modeling profile as the data source to create a model.

Once you build a model, you must move the resulting model file to the model server (this can be your PeopleSoft Process Scheduler server or a dedicated server). You must also define the model within PeopleSoft Customer Behavior Modeling. After you create a Customer Behavior Modeling model definition, assign it to a preconfigured model deployer. Then, the data mining model is available via the deployer in interactive mode.

Note. You must license the data mining tool.

To use data mining integration:

1. Set up third-party software and configuration.

   See the PeopleSoft Enterprise Performance Installation documentation for this release on PeopleSoft Customer Connection.

   1. Install the application on your PeopleSoft Process Scheduler server (default) or a dedicated server.

   2. Make various application server and PeopleSoft Process Scheduler configuration changes.

2. Configure one or more scoring engines.

3. Configure one or more model deployers.

4. Configure one model agent.

5. Create the predictive model:

   1. Create a PeopleSoft Customer Behavior Modeling profile to use as the data source for model creation.

   2. Export the profile to a text file and import it into your data mining application.

   3. Build the model from within the data mining application and save it.

   4. Move the model into the model directory that you defined when you installed the vendor applications.

      This is the same model directory as the one that you define on the Scoring Engine page.

   5. Define the model on the Data Mining Model Definition page.

   6. Assign the model to a preconfigured model deployer.

6. Score the customer records based on the model that the data mining tool generates.

   You score the records via:

   1. Interactive scoring.

      Use the Interactive Scoring page to test your configuration. To use interactive scoring from another PeopleSoft application, you must configure the provided API. To implement the API, see the Data Mining Integration document on PeopleSoft Customer Connection. You can also implement the Interactive Scoring API from a non-PeopleSoft application using XML.

   2. Batch scoring.

      Export the profile that you created to the data mining model for batch scoring. (The scores are imported automatically.)

Engines and Tables Used

This table lists the jobstream IDs, engines, and tables used during processing:

Note. The jobstreams that are listed in this table are for the sample data that PeopleSoft delivers. You can create your own jobstreams.

See Also

Using PeopleSoft Customer Behavior Modeling File Management

Exporting Data

Configuring and Using Scoring Engines, Model Deployers, and Model Agents

To configure and use scoring engines, model deployers, and model agents, use the Scoring Engine Definition (CR_DMI_PSERVER), Start Scoring Engine (CR_DMI_PSRV_STATUS), Model Deployer Definition (CR_DMI_MSERVER), and Start/Stop Model Agent — Batch Scoring (CR_DMI_MDL_AGENT) components.

This section provides an overview of predictive model deployment and discusses how to:

• Configure scoring engines.

• Use scoring engines.

• Configure and use model deployers.

• Configure and use model agents.

Understanding Predictive Model Deployment

Use two different types of servers—the scoring engine and model deployer—to deploy the predictive model.

• The model deployer accepts incoming requests to score one or more records and passes this request to the next available scoring engine.

• The scoring engine computes the scores and passes the results back to the model deployer.

• The model deployer ends the transaction by sending the results back to the client.

Pages Used to Configure Scoring Engines, Model Deployers, and Model Agents

Configuring Scoring Engines

Access the Scoring Engine page.

To configure the scoring engine:

1. Enter a description.

2. Enter an address and port.

   This defines the machine and port on which the scoring engine listens. This machine must have PeopleSoft Process Scheduler running in order to start the scoring engine from the Start/Stop Scoring Engine page.

   Important! Define addresses and ports in advance with your information technology (IT) department.

3. In the Specify Model Dir (specify model directory) field, select the check box and enter the path of the data mining model directory.

   Define this directory in advance. The path must be relative to PeopleSoft Process Scheduler, not your local workstation.

4. Enter the thread count (number of concurrent connections) that the scoring engine supports.

5. Enter the name and location of the configuration file that the system generates when you click the Write File button.

   You can give the configuration file any name. In a default setup, this configuration file lives in the same directory as the data mining vendor’s PeopleSoft deployer integration code on PeopleSoft Process Scheduler. The path is relative to PeopleSoft Process Scheduler. However, you must write the file and then manually place it into the location that you specify on this page.

6. Select Use Default Server or select Choose Server and enter an available server.

   This specifies the PeopleSoft Process Scheduler that you will use to start the scoring engine.

   Note. Define this server in advance with your IT department.

7. Enter the name and location of the working directory.

   In a default setup, this is the directory in which you installed the vendor’s PeopleSoft deployer integration code. Specifically, it is the directory in which ps_angoss.dll is installed on PeopleSoft Process Scheduler. Again, the path is relative to PeopleSoft Process Scheduler.

   Note. Define this directory in advance with your IT department.

8. Enter any comments in the Descr (description) field.

9. Save your entries.

10. Click Write File to generate the scoring engine configuration file.

Note. When the configuration file is generated, you must manually transfer (via file transfer protocol [FTP]) or save it to the desired configuration file directory location.

Using Scoring Engines

Access the Engine Status page.

Select an available engine from the search page; the Engine Status page appears. Use this page to manage the predefined scoring engines that you set up on the Scoring Engine page. You must start one or more scoring engines to use interactive scoring. The scoring engine is where the incoming connections from the web server are routed.

Click Start to begin the selected scoring engine. This sends the process to PeopleSoft Process Scheduler.

Configuring and Using Model Deployers

Access the Model Deployer page.

Deployers route incoming connections to available scoring engines.

To configure model deployers:

1. Enter a brief description of the model deployer.

2. Enter the URL to the deployer servlet directory.

   This is used on the Interactive Scoring page to test the configuration and is also used if you make interactive scoring available from other applications from the provided PeopleCode or XML APIs. In a default installation, the URL consists of the web server's HTTP address plus /servlets/ScoreServer/.

   Note. Define this URL in advance with your IT department.

3. Enter the name and location of the configuration file that the system generates when you click the Write File button.

   You can choose any name for the configuration file. The path is from the perspective of the application server. When the configuration file is written, you must move it manually to the web server, and the web server's weblogic properties file must be updated to point to this configuration file. Generally, you do this once in coordination with your IT department.

4. Enter the name and location of the feedback file.

   This file contains the log of success and fail results from the interactive scoring that is applied in your transactional application (for example, the call center) if you implement this feature via the Interactive Scoring API.

5. Enter any comments in the Notes field.

6. Enter the predefined scoring engines.

   Note that you can point the model deployer to one or more scoring engines. Enter the number of threads that the deployer can use; this number cannot exceed the number that you specified in the scoring engine definition.

7. Save your entries.

8. Click Write File to generate the model deployer configuration file.

   Note. You can create or edit this configuration file manually—the structure is simple.

Configuring and Using Model Agents

Access the Start/Stop Model Agent page.

To configure and use model agents:

1. Enter a brief description of the model agent.

2. In the Network Address field, enter the IP address or name of the machine that is to host the model agent.

   Generally, this is PeopleSoft Process Scheduler or a machine that hosts the scoring engines.

3. In the TCP Port field, enter a port number where the model agent listens for connections.

4. In the Server group box, select Use Default Server, or select Server Name and enter a server name.

   The server that you enter here starts and stops the model agent.

5. In the Working Directory field, enter the path to where ps_angoss integration libraries are stored.

   Generally, this is in the same directory where your data mining software is installed. The path is relative to PeopleSoft Process Scheduler.

6. Enter any comments in the Notes field.

7. Click Start to submit the process that starts the model agent.

   Click Stop to stop the model agent, and click Ping to confirm that the model agent has started.

Note. When the configuration file is generated, you must manually transfer (via FTP) it to the servlet classes directory on the web server.

See Also

Setting Up Interactive Scoring

Defining Models

To define models, use the Model Definition (CR_DMI_MDL_DEF) component.

Generally, after you configure the scoring engine and model deployer, you can create models at will and assign them to predefined deployers. It’s best to have one or more deployers waiting for models to be pointed at them.

This section discusses how to define models.

Page Used to Define Models

Defining Models

Access the Model Definition page.

Note. The following steps assume that you have already created a model and you want to deploy that model from PeopleSoft Customer Behavior Modeling.

To define models:

1. Enter a brief description of the model.

2. Click the Attach button to access a browser and select the file name and location of the model that the data mining tool created.

   This directory, which is relative to PeopleSoft Process Scheduler, must match the directory on the Scoring Engine page. In a default installation, the model files live in the same directory as (or a subdirectory of) the vendor’s deployer integration code.

3. Enter a predefined deployer to assign the model to it.

   This is required only for interactive scoring, not for batch scoring.

4. Select the model in the Dimension field.

   This value must match the dimension on the PeopleSoft Customer Behavior Modeling profile used to create the model in the data mining application. Generally, this value is Customer. Values are Account, Business_Unit, Campaign, Channel, Customer, Customermaster, Ded_Status, Department, Employee, Job, KPI, Location, Model, Operating_Unit, Product, Project, Project_Request, Scenario, Scorecard, Territory, and Vendor.

   The system uses the dimension to relate the batch scores to an object ID when importing them into the PeopleSoft Enterprise Warehouse batch scores table (CR_DM_SCORE).

5. Enter any comments in the Notes field.

6. Expand the Import Fields scroll area and select a model agent.

7. Click the Import button to automatically import the model's input and output fields.

8. (Optional) Select the field type for each field.

   Values are Char, DTTM (date and time), Date, Number, Time, and Unknown.

9. (Optional) Enter relevant comments for any of the fields.

   For example, comments in input fields could have possible ranges of values for each field, especially if there are fixed values. The input fields correspond to the relevant fields that are used in the profile to build the model.

10. Save your entries.

Note. You can assign multiple models to the same deployer.

Setting Up Interactive Scoring

To set up interactive scoring, use the Interactive Scoring (CR_DMI_ADHOC_SCORE) component.

This section provides an overview of interactive scoring and discusses how to set up interactive scoring.

Understanding Interactive Scoring

After the data mining tool creates a model and you save it in your working directory, you can configure the realtime model deployer and test the configuration from an interactive scoring page. You can rescore models on demand, using the most up-to-date information. Generally, use the Interactive Scoring page to check that the realtime scoring tools are configured properly. To use realtime scoring in a production environment, you should configure an interface to the scoring engine using the integration tools that PeopleSoft provides.

A PeopleSoft client application (such as PeopleSoft Customer Relationship Management) can access the predefined models to score records on an interactive basis, through a model-independent PeopleCode function. For example, a call center analyst can access a page that mirrors the PeopleSoft Customer Behavior Modeling Interactive Scoring page. The analyst can score the customer and, based on the score, cross-sell a product.

A third-party client application can access predefined models to score records on a realtime basis through an XML interface over HTTP.

Page Used to Set Up Interactive Scoring

Setting Up Interactive Scoring

Access the Interactive Scoring page.

The system displays the fields that correspond to the columns in the selected model (yet only the input fields that the model actually uses). Enter values for the customer record that you are scoring. (You must enter valid data into these fields.) The data is based on the profile data that is used to build the model. Some fields may be a specific range of values, and some may be any numeric value. You will know this from the data that is used to build the model. To make this data easily accessible in a production environment, you can export the entire profile in a text file and import it into your production environment.

Viewing Batch Scores

You can use the data mining tool to score customer records in batch, then import the results into PeopleSoft Enterprise Warehouse. Use the profile export feature to generate the scores, and the import feature automatically loads the batch scores into PeopleSoft Enterprise Warehouse.

This section discusses how to view scores.

See Exporting Data.

See Importing Data.

Page Used to View Batch Scores