Dissertation > Excellent graduate degree dissertation topics show

The Research of Topic Based Multi-document Summarization

Author: YueDaPeng
Tutor: WangTing
School: National University of Defense Science and Technology
Course: Computer Science and Technology
Keywords: Multi - document summarization Topic Natural Language Processing News Topic Detection and Tracking
CLC: TP391.1
Type: Master's thesis
Year: 2011
Downloads: 25
Quote: 0
Read: Download Dissertation

Abstract

The rapid development of Internet technology, the existing literature and knowledge to an exponential growth. Multi-document summarization technology can help people get important information from the large amounts of text, and can effectively reduce a great deal of time and effort spent by the user to read a very important value in this era of information explosion. At present, the news reports are often based on the form of the topic to start an event as a primer, and a series of associated or similar events reported organizations together to show in front of the reader. Topic-based document organization to clearly explain the causes and consequences of a series of news events and the ins and outs of a user-friendly query and read, and therefore by everyone welcomed and widely used. The research for this set of topic-based multi-document summarization. And compared to the ordinary set of documents, the high degree of duplication of information, topic-based set of documents irrelevant information, document content between closely. Extract Digest, if they can take full advantage of the nature of these do not have a general set of documents, you can get a better abstracts on the topic-based documentation set. This article looks at the topics based on the topic of the document set characteristics to improve ordinary digest algorithm. Improve their work, two points: to distinguish between the events to treat seed events and non-seed, give full consideration to the time attribute the summary sentences extraction and organizational. In achieving a topic-based news reports for processing objects, proposed and implemented on the basis and framework of the MMR (maximal marginal relevance) Abstracts extraction algorithm based on the topic of multi-document summarization. Extract topic keywords from the document set, taking into account the different events in which the removal of Abstracts status of seed events and non-seed, so handle both events. During sentence similarity comparison, taking into account the characteristics of time-sensitive news corpus, given a certain amount of time attributes of each sentence, and therefore a measure of the time to calculate the similarity between sentences. Sort of summary sentences, sentence time attributes and organizational structure designed for two different documents of a different sort. This paper and experimental evaluation of these abstracts methods TDT4 corpus of news reports, will compare the topic-based summarization system and two baseline Digest, achieved good experimental results.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Text Processing
© 2012 www.DissertationTopic.Net Mobile