| A new approach to interactive information filtering is presented: incremental Bayesian inference is applied to a multinomial model of text-document relevance as a means of learning user information needs over extended periods of time, through interactive data sampling. An information filtering agent acts autonomously on a user's behalf by filtering on-line document streams for text documents that are relevant to the user's information need and forwarding such documents to the user. For each forwarded document, the user is prompted to confirm or deny its relevance; a filtering agent that is able to incorporate such user feedback into its decision process can significantly improve its future document selection accuracy.; In contrast to nonprobabilistic information filtering models, which are based on heuristics and ad hoc techniques, the proposed probabilistic model provides a theoretical foundation for interactive information filtering. During empirical trials, the proposed probabilistic model has shown improved performance for certain measures, relative to nonprobabilistic models. |
