Research On Generation Method Of Evolutionary Multi-document Summarization Based On Sub-topic Enhancement

People's life has been changed greatly because of the rapid development of the Internet,our life are filled with all kinds of information and a variety of media information are increasing greatly everyday such as texts,images,audios,videos and so on.With the evolution of time,the relevant media information is evolving and updating constantly.There is no doubt that it is really difficult for a user to acquire the interested information completely and efficiently.So under such background,text summarization has been generated with the opportunity and developed into a hot spot in the field of natural language processing.Text summarization can be divided into two part,one is the traditional static text summarization and the other is dynamic text summarization that taking into account the time dimension.The traditional static text summarization generates the summarization which just aims at the media information that has the same topic and the same timestamp.However,with the continuous development and evolution of media information,the information has the feature of dynamic evolution obviously.At this time,because of the failure of meeting the needs of the general public of the traditional static text summarization,the dynamic evolution summarization has generated.The dynamic evolutionary abstract that introduces the time dimension bases on the static abstract.It aims to generate the summarization of the media information at different stages of development under the same theme.It not only need to consider the topic relevance and the low redundancy of the summarization but also need to consider the coherence and novelty of the summarization,so that the abstract of corresponding theme can be dynamic evolutionally with the development of time.In this paper,we propose a new multi-document evolutionary summary generation method based on subtopic enhancement,this approach considers both the relationship between sentences and the influence of the subtopic of each time period on the sentences,and to make the score of the sentences higher which is more related to the important subtopic,we sort the sentence comprehensively in subtopic level through the mutual reinforcement of the sentence and the subtopic.It is feasible in theory,and it also be verified the feasibility by the experiment results on the public dataset Timelines 17.In this paper,we also present an evolutionary multi-document summarization system.This system can grasp the news of Sina home page which users interested in online.it also can generate the evolution summarization for every news and user can read these evolution summarization to learn the evolution process of news.