Abstract
A paper filtering system that supports the effective collection of related technical papers is becoming important as the technological progress has been rapid. Two requirements for the paper filtering system are [A] the reduction of user's workload in specifying filtering conditions and [B] sufficient filtering accuracy. In this paper, we propose a paper filtering method that meets both [A] and [B] simultaneously by focusing on feartures of co-author research group, subject category, and terminology. The result of evaluation using 3,600 domestic society papers shows that the proposed methods improved mean average precision from 0.39 to 0.50 by 0.11 comparing with conventional pseudo relevance feedback method, which becomes closer to practical use.
