Collaborative Filtering by Personality Diagnosis: A Hybrid Memory- and Model-Based Approach

David M. Pennock

Artificial Intelligence Lab
University of Michigan
Ann Arbor, Michigan 48109-2110

Eric Horvitz

Decision Theory & Adaptive Systems Group
Microsoft Research
Redmond, Washington 98052-6399

Access postscript or pdf file.


The growth of Internet commerce has stimulated the use of collaborative filtering (CF) algorithms as recommender systems. Such systems leverage knowledge about the known preferences of multiple users to recommend items of interest to other users. CF methods have been harnessed to make recommendations about such items as web pages, movies, books, and toys. Researchers have proposed many approaches for generating recommendations. We describe and evaluate a new method called personality diagnosis (PD). Given a user's preferences for some items, we compute the probability that he or she is of the same "personality type" as other users, and, in turn, the probability that he or she will like new items. PD retains some of the advantages of traditional similarity-weighting CF approaches in that all data is brought to bear on each prediction and new data can be added easily and incrementally. Additionally, PD has a meaningful probabilistic interpretation, which may be leveraged to justify, explain, and augment results. We show empirically that PD provides better predictions that all four of the algorithms tested by Breese et al. [1998] on the EachMovie database of movie ratings. The probabilistic framework naturally supports a variety of descriptive measurements---in particular, we briefly consider the applicability of a value of information (VOI) computation.

Keywords: Recommender systems, collaborative filtering, agents, diagnosis of preferences, probability, decision theory.

In: IJCAI Workshop on Machine Learning for Information Filtering, International Joint Conference on Artificial Intelligence (IJCAI-99), August 1999, Stockholm, Sweden.

Author Email:,