The Commerce Server Predictor resource is a powerful data mining tool that enables you to provide predictive capabilities for your Web site, for example, to display product recommendations. It also enables you to analyze the characteristics of the users visiting your site, and discover relationships among those characteristics. You can then use this information to target content to users who have similar characteristics.
The Predictor resource mines data collected in the Data Warehouse. The Data Warehouse database can store all site data, including user profile data, click-history data, and transaction data. Although the Data Warehouse database is quite large, the capabilities of the Predictor resource can efficiently mine the data.
Using Prediction in Web Site Management
Your site developer uses the Predictor resource in Commerce Server to add predictive capabilities to your site. The Predictor resource enables you to perform the following profiling and targeting activities:
The Predictor Resource builds analysis models from data. An analysis model is a set of statistical relationships based on known properties of past site users, their purchase history, click history, or other behavior. The model contains information about the types of users who visit your site, but it does not contain information about specific users. The detailed information used to create an analysis model is stored in the Commerce Server 2000 Data Warehouse, or another data source.
Analysis models enable you to perform implicit profiling and to target content to users:
Two types of analysis models are used in Commerce Server: Prediction models and Segment models.
You use a Prediction model, a collection of decision trees also known as a dependency network, to provide real-time purchase recommendations to users visiting your site, and to guess unknown profile properties about users. A Prediction model summarizes relationships in the data in the form of rules. For example, a Prediction model may say that if a visitor to your site is male, over 55, and purchases sports clothes, then he is also likely to purchase golf equipment. You can use this model to make real-time recommendations for golf equipment to users who match this profile.
Prediction models typically provide recommendations that are more accurate than human-generated rules, as they predict based upon the previous activity on the site; consequently, they usually result in more sales.
To analyze Prediction models, you use the Prediction Model Viewer in Commerce Server. For information about viewing Prediction models, contact your system administrator.
A Segment model, also known as a cluster model, partitions users who tend to have similar properties or behavior into segments. You can use these segments to gain an understanding of the users who visit your site and also for subsequent marketing. For example, you notice that sales have skyrocketed for a popular fantasy book for children. You discover that most of the purchases of that book are made by users in a segment with the following characteristics: female, over 40, income above $50,000 per year, and college educated. You can now offer similar products to users in this segment.
You use the Segment Viewer module in Business Desk to analyze Segment models. For information about the Segment Viewer, see Analyzing Population Segments.
Each analysis model is based on a model configuration, which is a description of the data to be used to build the model. A model configuration specifies:
Commerce Server includes the transactions model configuration, which is defined in the Data Warehouse schema. You use the transactions model configuration to build models based upon purchase history. If your site requires a custom model configuration, your site developer can build a new one.
As soon as you have collected data, you can use the model configurations to build analysis models. To build an analysis model, the system administrator uses the Predictor resource in Commerce Server Manager. Models are not built on the Web server, so the building process does not impact the performance of your site.
After the model is built, the system administrator can move a copy of the model to each of your Web servers (all Web servers use the same model), and then enable predictive capabilities on your site. You can view the Prediction model or the Segment model to analyze the patterns of the users visiting your site.
After you have gathered a significant amount of new data, your system administrator should rebuild your analysis models. You build analysis model configurations infrequently, however, only when you change what you are trying to predict or analyze.