Prediction and Data Mining
The Commerce Server Predictor resource is a powerful data-mining tool that you use to to provide predictive capabilities for your Web site, for example, to display product recommendations. You can also use the Predictor resource to analyze the characteristics of the users visiting your site, and discover relationships among the 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
Analysis Models
Creating Analysis Models
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:
Provide intelligent cross-sell support. Suggest items that a user may be interested in by correlating properties about the user, or items the user has ordered, with a database of items that other users have ordered previously. You can restrict the items that are suggested to a particular part of the catalog.
Segment users based on their behavior or profile. Analyze these segment models to discover the characteristics that partition those users into population segments. Your system administrator builds segment models that you can analyze using the Segment Viewer module in Commerce Server Business Desk.
Fill in missing values in your user profile data. Predict a missing value, for example, a user profile property that the user did not supply.
Analysis Models
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 Data Warehouse, or another data source.
You can use analysis models to perform implicit profiling and to target content to users:
Implicit profiling. If profile information about a user is missing — for example, the gender of a user — this information can be extrapolated by the analysis model based on the aggregate properties of the entire user population who visits your site.
Targeting content. You can use analysis models to add predictive capabilities to your Web site. For example, you may want to present users with a ranked list of recommendations for products in your catalog or to display advertisements that may be of interest. You can also view analysis models to analyze the characteristics of the users visiting your site, discover relationships among the characteristics, and then target content to users who have similar characteristics.
Two types of analysis models are used in Commerce Server: Prediction models and Segment models.
Prediction 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, see Viewing Analysis Models.
Segment Models
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 users in a segment with the following characteristics make most of the purchases of that book: 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.
Creating Analysis Models
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:
A data source for building the analysis model. Typically the data source is the Data Warehouse since it contains vast amounts of user data. However, you can use an external database to build an analysis model.
Attributes for the model. An attribute identifies which columns of user data (represented in a table) should be included in the model. An attribute may include user profile data, purchase history, or click history.
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, a site developer can build a new one. For information about creating custom model configurations, see Predictor Schema.
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.