Optimal Queries in Information Filtering

Information filtering has become an important component of modern information systems due to significant increase in its applications. The objective of an information filtering is to classify/categorize documents as they arrive into the system. In this paper, we investigate an information filtering method based on steepest descent induction algorithm combined with a two-level preference relation on user ranking. The performance of the proposed algorithm is experimentally evaluated. The experiments are conducted using Reuters-21578 data collection. A microaverage breakeven effectiveness measure is used for performance evaluation. The best size of negative data employed in the training set is empirically determined and the effect of Rnorm factor on the learning process is evaluated. Finally, we demonstrate effectiveness of proposed method by comparing experimental results to other inductive methods.