A Korean Information Retrieval Model Alleviating Syntactic Term Mismatches

In Korean information retrieval, term mismatches between indexing terms and query terms are a serious obstacle to the enhancement of retrieval performace. Term matches are not produced because of a space usage of compound nouns and also various representations of a phrase. This paper presents an extended model of Korean information retrieval of alleviating these term mismatches between indexing terms and query terms. In this model, we segments compound nouns into unit nouns by using statistical information and a preference rule. And then we synthesize unit nouns or single nouns into synthesized compound nouns by adjacency constriants. Among all candidates of synthesized compound nouns, we lter out meaningless compound nouns by mutual information and the relative frequency of category pairs. Moreover, we perform similarity computation considering partial matching for compound nouns. The experimental results show that the proposed method can overcome the diierence between surface forms of terms.