A Study Of Automatic Summarization For English Document By Citing Sentences

Quickly moving to a new area of research is painful for researchers due to the vast amount of scienti?c literature in each ?eld of study. One possible way to overcome this problem is to read the summary of the topic. From the summary,one can get the backgroud,the current stat and the future of the topic.However,it is not easy to find a good summary of a topic,because it must be written by an expert of that topic,and it is quite time consuming.So,nowadays,more and more researcher are getting to study automatic summarization of research topic. An ideal such system will receive a topic of research, as the user query,and will return a summary of related work on that topic.As an important step of this goal, automatic summarization of documents has been wildly studied. Studies have shown that different citations to the same article often focus on different aspects of that article. Hence through analyze the collection of citations,one can get the main contributions of a paper and how that paper affects others.In this paper, first introduce a method to summarize an article using citations,and then describe the shortcomings of this method,finally improve this method by: 1.Add the compare information into the summary.The summary generated by the original method contains only contribution information,which describe what the author had done,but ignore the compare information,from which one can catch the chronological order and the progress in that particular ?eld of study.Compare information is very important to generate the summary of one topic,so this paper add the compare information into the summary. 2.Make the summarization contain more contribution information and more important contribution information.The final summary’s quality depends on the clusting precision.This paper use a different cluster methon from the original,and determine the patameter by experiments.And show the results of five natural language processing aspects. 3.Use more accurate method to compute the weight of the citing sentences.After clustering,it has to select the most salient sentences from every class to form the summary.So it has to compute the weight of citing sentences.The original method only use the similarity of text,except this,this paper also consider the citited by number and the corresponding author’s influence factor of the citation paper to improve the accurace of weight computing. 4.Improve the evaluation method to make it more just.In the original method,the evaluation standard is depend on the experiment results.So it is injustice more or less.This paper improve the evaluation method to make it more just.