| The explosion of the World Wide Web has created a demand for new ways of managing dynamically changing information. Query-focused multi-document summarization requires summarizers to piece together information from multiple documents to answer a question or questions. It will meet the growing needs of people for information. Because query-focus multi-document summarization considered both query information and themes of relevant document set.The research can be summarized as follows:Firstly, for query-focus multi-document summarization, this paper proposed the strategy of summary sentence selection using keywords extraction. It reckoned the word's query related feature through the technology of query expansion, and obtained the topic related feature through likelihood ratio, then combined the two features to estimate the importance of every word. The score of candidate sentence was the sum of the importance of words it contains, and the modified MMR technology was adopted to generate the final summary. This paper introduced features fusion into the word level, can be used to describe precise information in thinner granularity.Secondly, this paper proposed the strategy of summary sentence ordering using clustering and template. Summary sentences are clustered into sub-topic set, this keeps the sentences in the same sub-topic set together. The template is selected by the summary representation of document, this makes the sentences are coherent logically. Sub-topics are ordered by the relational position in template, the sentences in the same sub-topic set are ordered by the position in template. The experimental result shows that the proposed method improved the readability of the summary.The experimental result shows, the strategy of summary sentence selection using keywords extraction and the strategy of summary sentence ordering using clustering and template are effective, can improve the performance of summary effectively. |
PDF Full Text Request