Query Expansion through Feedback in Japanese Information Filtering Based on the Probabilistic Model

摘要

This paper reports experiments in query expansion through relevance feedback and local jeedback for online Japanese news filtering with the probabilistic NEAT system. The expansion terms are extracted from either the full texts or the headings of the relevant documents Using the standard BMIR-J1 test collection and a separate collection, relevance feedback is shown to produce an improvement of up to 18/100 in average performance. Local feedback is shown to give a smaller, but still significant, improvement for the BMIR-J1 and BMIR-J2 collections. In addition, using the query complexity groups defined for the collections, we compare our relevance and local feedback results.