Information retrieval systems--which return information according to a textual description and user queries--have been around for quite some time. Their natural evolution is recommender systems, which return information automatically based on user preferences. This paper describes just that: a recommender system built as an extension to existing concepts about information retrieval systems.

The system described here constructs relationships between users and items based on concepts widely used in information retrieval: precision and recall. Precision is the probability of a given user being interested in the selected item, and recall is the probability of selecting a user interested in a given item. The relationships thus built create a belief network, or a common space of users and items represented in terms of explicit and implicit factors.

The paper starts by describing a basic probabilistic inference model composed of users, items, and propositions, or information associated with users. The strategies used to associate information with users can be either item oriented (find users interested in items) or user oriented (find items interesting for users); in any case, these are rather common strategies found in existing information retrieval systems. Then, the paper extends this model using probabilistic inference: items and users are paired based on previous user experience, which is defined with the HITS method; the ranking of item-user pairing probabilities is performed via explicit and implicit factors.

Next, the paper tests the system on two datasets: the outdoor movements of people in Beijing, and results from a movie rating application. These tests measure precision and recall, as defined above, and give results in terms of explicit and implicit factors. The paper concludes with recommendations for future work: the model should be improved by considering more factors, namely social and geographical ones.

This is a research paper from people working in an academic environment; that is, it is full of mathematical formulae. It is meant to show the advancement in the field by this group, and to foster further advancements within the whole community.