Personalized information delivery: an analysis of information filtering methods

Peter W. Foltz, Susan T. Dumais
1992 Communications of the ACM  
This research tested methods for predicting which Technical Memos (TMs) best match people's technical interests. Within Bellcore, approximately 150 new TMs are published each month, yet very few are relevant to any single person's interests. In a six month study, 34 Bellcore employees were sent monthly personalized lists of new TM abstracts that were predicted to best match their interests. These predictions were made using two methods for describing technical interests; one based on sets of
more » ... words that the employees provided, and the other using feedback about previous abstracts they found relevant. Two information retrieval methods were tested to make the predictions, one using standard keyword matching, and the other using Latent Semantic Indexing (LSI). All four methods effectively selected relevant abstracts. The best method for filtering used LSI with feedback about previous relevant abstracts. Feedback using previous relevant abstracts provided an efficient and simple way of modeling people's interests. Overall, the filtering methods show promise for presenting personalized information.
doi:10.1145/138859.138866 fatcat:di3zxdi5v5bh7jvkiqwrfm3gm4