Extracting Concept Hyponymy Relations And Multifaceted Definitions From Books

Nowadays, knowledge base is increasing crucial for text understanding,but it is still very challenging to build a largescale knowledge base for arbitrary domains. Fortunately,massive books have been digitized in many digital libraries,and people can learn knowledge from books. If we can learn hyponymy relations and definitions from massive books directly,it would be very helpful to build knowledge base.In this paper, we proposed a novel approach to extract concept hyponymy relations and multifaceted definition from massive books. The hyponymy relations and coordinate terms are extracted from book catalogs through concept validation and filtering conditions, and then a taxonomy can be built from the hyponymy relations and coordinate terms. Meanwhile, definition candidates can be extracted from books by catalog based search engine, and then the multifaceted definition can be generated by clustering candidates and ranking the representative ones.With this approach, we can build the taxonomy entirely from scratch. The practice in CADAL digital library and the experiments demonstrate the feasibility of our approach.