Interest Areas

Interest areas attempt to quantify the user behavior across client's website (sometimes different domains) to determine what content or site sections visitors are most interested in.

Automated decisioning and targeting algorithms base their prediction on variables that are stored in the user profile. A profile can contain two types of variables: continuous and nominal values. Thus, the quality of predicting the best offer (modeling group) to display relies heavily on the number and quality of data (variables) from the user profile. Currently, there are no systemic variables in user profile that would describe the user behavior on the site, as most of them are simple characteristics such as geo location or browser type.

The rationale behind tracking user activity on the website is related to the fact that different users have different intentions and interests when visiting a site. Because most of the clients' websites are very diverse and often one site comprises tens, if not hundreds of different compartments, users tend to focus their browsing on isolated areas. This tendency of visiting isolated areas of interest is often predictive of a user's temporary behavior.

The quantification of interest areas takes the form of one or more continuous variables stored and updated in the user profile on each request. Each variable represents the degree of interest the visitor expresses for a particular area of the site.

This information is often very predictive in determining what content is best to show. Site URLs are mapped to higher-level categories, or "interest areas" (for example, mapping men's product page viewing to a "men's" interest area), without need for manual updates or monitoring. Client sites are automatically crawled based on the domains that send mbox calls, and URLs are grouped based on the site's structure, as defined by the site's site map or general URL structure. Very common pages, like the home page, are removed from the interest area groupings.

Interest areas are stored in the training data and, if they are predictive, affect the model and display in the insights report. This is all done automatically. No manual edits or updates to interest areas are needed.