Research On Event-oriented Automatic Summarization

In recent years, the automatic summarization is an important branch of natural language processing, which becomes more and more famous, and the concept of event has gradually been adopted in the field of natural language processing, the event as a knowledge unit in line with the laws of human understanding of the world. So, this paper studies for the generation of automatic summarization oriented on event, which means combining with event-oriented research, and making use of automatic summarization technologies to get a summary. We mainly can make two parts to expand this paper.1) Researching on event elements recognition. This paper mainly proposes to identify the elements of event according to the event formal definition of six groups. By observing the text actually, we find that we can only get four kinds of elements from text such as event denoter、event object element、event time element and event place phrase. So, we will focus on these four kinds of elements’ language performance rules to propose respective identifiable methods. We will propose a method which combining with expanded trigger list and multi-feature fusion machine learning to recognize event denoter. To recognize event object element, we firstly propose to complete the deletion of event object, then filtering the objects according to the language performance rules, and lastly using the co-reference resolution to make it true. Recognition for event time element, we propose to identify time expression and calculate event timing relationship. For event place phrase recognition, we combine maximum entropy with rules to identify it. Finally, we test them on CEC corpus, the results show that the proposed methods all achieve some certain effect.2) Automatic summarization research based on event elements. We consider that though the tests of first study get quite certain results, they still have something bad. As we must improve the rigor of scientific research, we need to avoid errors being cascaded. So, this paper still uses the CEC corpus as the given event elements to research automatic summarization. It mainly studies for the automatic summarization based on event elements from the graph construction and calculation. Firstly obtaining the event elements through the tagged CEC corpus, and building an event element undirected graph, then calculating nodes’ and edges’ weights of the undirected graph, finally getting the concise summary sentences by the giving compression ratio and outputting the text summarization in accordance with the original text sequence. The test results show that the method being proposed in this paper gets quite good results and the average value F relatively ideal as well.