Experiments on the TREC-8 Filtering Track

For this year's TREC, KDD R&D Laboratories focused on the adaptive ltering experiments of the Filtering Track. The main focus of our research was the development and evaluation of the pro le updating algorithm. Our pro le updating algorithm is based on the query expansion method based on word contribution[1][2]. Given manual feedback, our QE method has achieved high performance in the ad hoc track. Therefore, our hypothesis is that this method should work well in the ltering track. We will describe how we implemented this method to the ltering track, and analyze experiments. Our o cially submitted results to TREC were generated from a system with a major bug. The results described in this notebook paper are based on the bugxed version of our system.