Research On Event Type Recognition

With the rapid development of information technology, network has gradually become a huge data source, and contains many valuable information. Some of these information are about the all kinds of events which already occured or is just happening. For example, activities of some countries’s governers, trend of firms and so on.Event extraction is a means of getting the information you need automatically from the all kinds of text types, including entity extraction and relationship extraction. It can be divided into two steps:(1) event type identification, in this step event type is confirmed, and sub-type (2) event elements identification, label out participants of entity and their roles of current type event.This essay is dedicated to improve the research of event type recognition. It’s aim is to resolve large scale data oriented problems, data with high complexity under the guarantees of results’accuracy, achieve helpful information’s fully mining in the text, and finally, enrich the content of event extraction.The current event type recognition system’s range is out of actual demand, either based on sentence level, or based on article level, the view is either to huge or too small; on the other hand, most research focused on studying elements of events and trigger words, and event relation recognition of them is based on judgement of all words in the text, typically, the high redundancy of pending sentences as well.These all caused serious burden to the machine, and the introduction of numerous counterexamples also cause extremely imbalance of positive and negative cases.This article aim at resovling existing problems about unaccuratly range of the event type recognition, high redundancy of sentence identification, and low reliability. The main works and contributions of the essay are as follows:1. Propose a chunk division method based on segmentation technology, it can specify rangement of text which is to be extracted between sentence level and discourse level, make every block contains a number of events with a same topic, at last make a well foundation for the next step.2. This article put forward a method of event type identification based on sentences filtering, it can effectively solve the problems of traditional event type identification which is about imbalances between positive and negative cases,and also realized high precision, enhances the adapt ability of event type recognition as well.First of all, according to well division of text block, filter out the false event sentences, as well as the factual information of them. The use the method of multiple knowledge fusion to represent candidates of event instances. Finally, using machine learning method SVM based on multivariate classification to realize the event type of event instances’candidates.