Research And Application On Chinese Topic Event Extraction

With the rapid development of society,prople are surrounded by the information ocean.Extracting event information accurately and effectively from a large amount of disorderly,messy,unstructured data is the goal of event extraction.Currently,most researches extract event information(state or action) in sentence scope.However,the information of a topic event is comprised of several states and actions,and distributed in multiple documents.Current event extraction method in the scope of sentence can not meet the requisition of topic event.This paper divides the topic event extraction into three scopes.Firstly,it extracts meta-event(state or action) in sentence scope.Secondly,it extracts topic event segment information in every document according to co-reference and the frame of topic event.At last,it can get a complete description of topic event,by combining the topic event segments from several documents.This paper firstly introduces the basic concepts of event extraction,the current research status and the main difficulties in event extraction.Secondly,it brings forward a extraction method,which map semantic role of the triggered verb to the corresponding event element to realize the extraction in the scope of sentence by the semantic role tagging technology.Thirdly,it proposes an event semantic description frame based meta-event to describe framework of a topic event.It standardizes the important event arguments including person,time and place after extracting all meta-events.And then it uses the relations between some important semantic vector(person,time and place) to combine the meta-events to a topic event segmen.Fourthly,it proposes a cross document topic event clustering method which is based on document summary.After clustering,a cluster is a set of documents about the same topic event.At last,it gets complete information of a topic event by integrating the topic event segments from the cluster.The experimental results in the visit topic event extraction show that the topic event semantic description framework based on meta-event is valid to explain the key information of a topic event,and the normalization and using semantic information are also very important for the extraction.In the experiment of topic event clustering,our clustering method based on document summary and the improved similarity calculation is motivated by experiment.