Research On Key Technology Of Open Domain Event Extraction

With a rapid development of Internet applications,the access to information on the network system online grows with a splitting speed.Now there's a particularly important task for researchers to seek really useful information from such a massive data set quickly and efficiently,and then there came the research of information extraction,in order to find the task a way out.As a sub-task of information extraction,the event extraction technology is based on the underlying natural language analysis technology,which depends on the result of processing techniques,such as word segmentation,POS tagging,named entity recognition and syntactic analysis.However beyond these technologies,the exploration of real meanings behind the sentences plays a even more important role,which requires far more great endeavor to pay a study.Most of the corpus used in traditional event extraction research are highly restricted to its domains as well as event categories,then followed by machine learning or pattern matching methods,in order to identify the event type and event elements.As a result,traditional event extraction systems perform poorly in terms of portability.In this paper,without necessary to specify the event type in advance,the content of the corpus extracted from the open domain event involves various fields.This paper also pays attention to open domain event extraction as two phases:(1)Open domain event trigger extraction.In the stage of triggering word extraction,this paper proposes a R-ME model extraction trigger based on the combination of rule and maximum entropy model,with manually develop rules applied among them.Advantage of this choice lies in the ability to accelerate the rate of characterizing and extraction,but with the disadvantage as poor performance in complex sentence extraction.Based on this context,this paper combines the above two methods,and works through experiments to verify the effectiveness of fusion methods.(2)Open domain event element extraction.this paper constructs a label analyzer,which integrates the linguistic features between event triggers and candidate event elements into the parser.The conditional random field model is used to identify the event elements,result of the experiments proves Good extraction effect.Finally,on the basis of the above two tasks,we design an open domain event extraction prototype system.