Study Of Chinese Event Information Extraction Based On Hownet Semantic Relation

With the development of the Internet technology and exponential increase of online text, the contradiction of"explosion of information and knowledge-hungry"grew more and more fiercely because of amounts of non- structured information. So how to automatically gain structured information from non- structured information text is an important research topic now, which is the reason we study event information extraction.As a high-level study task of information extraction, event extraction evaluation was unsatisfactory, and especially the Chinese event extraction, there was larger conflict between basic research and mission requirement on account of the basis of Chinese information processing is still failed to achieve the semantic analysis. Current research is still at the stage of continuous improvement, and there are no mature available systems.Based on the analysis of the pros and cons of the two methods, pattern matching and machine learning combining method was used to study Chinese event information extraction. The basic idea of the event information extraction methods is in accordance with the basic process of pattern matching, and semantic role labeling during the event and event specific information is filled with statistical methods used in the course.The Hownet semantic relations between the information and knowledge resources are introduced in this paper to identify the semantic association and solve the problem of semantic missing. Pattern-matching methods in terms HowNet-based similarity calculation are used to study the Chinese named entity recognition process, which includes the basic information of the event (including the nature of the incident, time, location,etc.). Experimental evaluation has achieved good results later. Semantic resources of Howet event concept class is used to conduct semantic annotation of events and address the difficulties of trigger word detection and role annotation. Finally the texts of catastrophic events collection is selected as the study object and layer-frame is used to indicate all kinds of side information, and then layer-type finite state machine approach is used in the extraction process of event information.