Internet recommendation systems

Several online firms, including Yahoo!, Amazon.com, and Movie Critic, recom mend documents and products to consumers. Typically, the recommendations ar e based on content and/or collaborative filtering methods. The authors exam ine the merits of these methods, suggest that preference models used in mar keting offer good alternatives, and describe a Bayesian preference model th at allows statistical integration of five types of information useful for m aking recommendations: a person's expressed preferences, preferences of oth er consumers, expert evaluations, item characteristics, and individual char acteristics. The proposed method accounts for not only preference heterogen eity across users but also unobserved product heterogeneity by introducing the interaction of unobserved product attributes with customer characterist ics. The authors describe estimation by means of Markov chain Monte Carlo m ethods and use the model with a large data set to recommend movies either w hen collaborative filtering methods are viable alternatives or when no reco mmendations can be made by these methods.