Combining Personalized Agents to Improve Content-Based Recommendations

Ratings-based recommender systems typically predict user preferences for items based on the users preference history, information about items, and the preferences of similar users. In content-based recommending, the similarities between items the user has previously expressed interest in form the basis for recommending new items. There are a number of practical reasons why users may not rate all of the items they have experience with, a fact that indicates ratings are not missing at random. We introduce a missing data model that takes this observation into account. By combining the personalized content models with missing data models, we build classifier agents for each user using the predicted ratings of the first two models. These staked agents use collaborative filtering to construct a hybrid recommender system that improves upon the baseline scores produced by the content-based recommender on popular movie ratings data set.