| Document Automatic Summarization is an important branch of Nature Language Understanding. Recently, as Internet develops, more and more information is available online, providing a huge amount of resources for people. It takes lots of time for people to browse the web for what they need People can get information by search engines. But search engines have some serious defects. For example, so lots of pages returned by them are the same or similar. That is why people cannot get information rapidly and effectively. The aim of Web Automatic Summarization is to solve such problem. It can provide people a full-information and concise web summary, and help people get information effectively.Sentence Extraction based Document Automatic Summarization is an old-line, efficient, and widely used method. This method extracts sentences from document through statistic method or heuristic method. Then these sentences are combined as a summary according to length regulation. In this paper, several key technologies of Based-on-multi-characters Chinese Web Automatic Summarization system are discussed, including web pre-process, keywords extraction and sentences weight calculation. Before summary analysis, we need pre-process the web document. In a HTML document, there are a lot of sentences which are foreign to the web content. We must filter out them. In this way, it reduces the scanning area when generating summary, furthermore it gets exacter content.Extracting keywords plays an important role in summarization. In this paper, we use frequent sequence mining method to find keywords. We use improved PAT-Tree technology to provide a structure for a document, and combining with net frequency method to mine the frequent sequence. The result of experiment proves that it is better to extract name, place-name, new words in webs, gerundial phrase and noun phrase by the sequential mining method.In tradition, it usually uses the linear interpolation method to calculate the weight of sentences. In this paper, we improve the traditional method by adding two affection factors. They are sentence significant factor and the sentence similarity with topic. In this way, we may obtain the sentences which are more correlative with document topic. The experiment shows that the selected sentences exactly reflect the document topic.This paper introduces a Chinese web Automatic Summarization system that is based on technologies mentioned above. The experiment indicates that the system is better than present Chinese Web Automatic Summarization on content estimation and summary readability. |
