Merging Case Relations into VSM to Improve Information Retrieval Precision

This paper presents an approach that merges case relations into the well-known Vector Space Model (VSM), leading to a new model named CVSM (Case relation-based VSM). A Chinese case system with 23 case relations is established, and a Chinese Olympic news corpus of 7,662 sentences, denoted COCS, is constructed by manual annotation with these 23 case relations. We use 50 queries on COCS as a test set. Experimental results on the test set show that C-VSM outperforms W-VSM (Word-based VSM) by 3.4% on the average 11-point precision. It is worth pointing out that almost all the previous studies on semantic IR obtained no better, even worse, results than W-VSM, our work thus validates the usefulness of case relations in IR through the validation is still preliminary. The proposed model is believed to be language-independent.