Research And Application Of Multi-document Automatic Summarization

Recently, with the popularity of Internet, more and more information is available online, providing huge a mount of resources for people. Now, the most important way for people to get information is by search engines. But search engines have some serious defects. For example, lots of relevant documents returned if you use search engines to get information, and lots of them are the same or similar. That is why people cannot get information rapidly and effectively. The aim of Multi-Document Automatic Summarization is to solve such problem. It can provide people a full information and concise document, and help people get information effectively.In this paper, several key techniques of Chinese Multi-Document Automatic Summarization are discussed. They are sentence similarity computation, sub-topic information identification and sentence optimal selection, summary sentence sorting, evaluation of multi-document summarization.First of all, the method of sentence similarity computing is studied deeply. In this paper, a method of sentence similarity computation that based on semantic dependency is discussed in detail. This paper proposes a novel approach for sub-topic segmentation based on maximum tree algorithm. Based on sentences similarity computing maximum tree, we combine the similar sentence into a combination, representing a sub-topic. A method of sentence optimal selection is proposed. In order to cover the general information in the case of limit space, we divide the extracting process into two phases: one is the sorting of the sub-topics, and the other is the sentence selecting of the sub-topics. We make a research of scoring strategy of the sub-topics, based on this, we study many kinds of the sorting methods of the sub-topics, and then make comparison of them. Make an optimal selection of the sentences of the sub-topics in order to maximize the information coverage ratio. The method of sentence sorting of the summary is proposed, and then find the best sorting method. At last, the method of abstract evaluation of multi-document automatic summarization is studied deeply.