Using syntactic information in handling natural language queries forextended boolean retrieval model

There are considerable evidences that trained users can achieve a good search eeectiveness through structured boolean queries rather than simple keyword queries because boolean operators can help to make more accurate representations of users' information search needs. However, it is not normally easy for ordinary users to construct eeective boolean queries using appropriate boolean operators. In this paper , we propose a syntax-based technique for handling natural language queris and phrases for extended boolean retrieval model in order to pursue both search eeectiveness and user convenience. First, natural language queries are syntactically analyzed using Korean natural language parser and the resulting syntactic trees are structurally simpliied using tree-simplifying mechanism in order to catch the logical relationships between keywords. Secondly , in a simpliied tree, plausible noun phrases are identiied and added into the tree as new additional keywords for more precise retrieval. Finally, the tree is automatically converted into a boolean query using some mapping rules and linguistic heuristics. We also propose an n-best tree method which uses top n syntactic trees to compensate for detrimental eeects of a single incorrect top syntactic tree. In the experiments using KTSET2.0 (Korean standard document set), we showed that the proposed method outperformed natural language models without any syntactic analysis by 23% and, surprisingly enough, outperformed even manually constructed boolean queries by 8% in the 11-point average precision measures.