A Chinese Atomic Event Extraction Method Based On Improved HMMs

Chinese event extraction is a subfield of Chinese information extraction,focusing on how to extract structured knowledge from unstructured text,and it is related to many techniques and fields,such as natural language processing,data mining,machine learning,database and so on.So far,it has been widely used in textual entailment recognition,information retrieval,stock price prediction,coreference resolution and community question answering.The traditional event extraction methods usually regard event recognition as a classification problem,mining events from text via machine learning methods or event templates,which can only solve the event extraction problem in certain fields.According to the definition of atomic event,atomic event is type-independent and can structure text more comprehensively.For atomic event extraction,this paper puts forward a type-independent model which is based on improved HMMs and solves event extraction as an annotation problem.The improved HMMs takes the transition and observation of current state and its historical state into account,and the relative position feature has also been taken into consideration.Moreover,error correction rules generated from K-means algorithm are utilized to improve the performance of the statistical model.To evaluate the performance of the proposed approach,this paper utilizes Precision,Recall and F1-score as standard.Experimental result shows that the improved HMMs model can extract atomic events from an unstructured text more effectively and efficiently.