A taxonomy fuzzy filtering approach

Our work proposes the use of topic taxonomies as part of a filtering language. Given a taxonomy, a classifier is trained for each one of its topics. The user is able to formulate logical rules combining the available topics, e.g. (Topic1 AND Topic2) OR Topic3, in order to filter related documents in a stream. Using the trained classifiers, every document in the stream is assigned a belief value of belonging to the topics of the filter. These belief values are then aggregated using logical operators to yield the belief to the filter. In our study, Support Vector Machines and Naïve Bayes classifiers were used to provide topic probabilities. Aggregation of topic probabilities based on fuzzy logic operators was found to improve filtering performance on the Renters text corpus, as compared to the use of their Boolean counterparts. Finally, we deployed a filtering system on the web using a sample taxonomy of the Open Directory Project