earticle

영어

This study aims to discuss the limit of actual keyword-based information retrieval systems and to present the finite-state Local Grammar Graphs(LGGs) constructed in this work for the purpose of the accurate recognition of queries by natural languages. We observed some real queries about the “agent(i.e. person & company)” in IT domain, such as “Who's the president of NHN Corp.?” “Let me know if Mr. Lee Jae-yong has a Ph.D.” “What's the latest model of Digital Cameras produced by Samsung?”. From this observation we classified the query types into 9 classes and divided them again into 38 sub-classes. Several graphs for each sub-class were built and we obtained finally 163 finite graphs that represent more than 4 millions of query patterns by natural language such as ‘-i gaibalha-n(develop-Sd) -neun muos(what)-ipni-gga?’(What's the LCD TV developed by Samsung?'). Empirical results show that our LGG graphs can recognize 90% of real queries by natural language and we notice that some types of complex phrases and adverbial sequences should be treated in our LGG patterns in the future works.