基于句子关系图的网页文本主题句抽取*

针对网页文本结构信息少、噪声大的特点，将句子看作点，将句子间的相似性看作边，用句子关系图描述文本中句子间的关系。抽取文本主题句的任务转化为搜索图中边最多的点。利用语义词典，将句子相似度定义为句子语义相似度，解决短文本词频相似度低的问题。选用互联网公开语料进行测试，抽取的主题句达到平均80.6%的可接受性。

Extracting Topic Sentences form Web Text Based on Sentence Relationship Map

Concerning the issues of Web text with little structure information and big noise, sentences are viewed as nodes and similarities between them are viewed as edges, a relationship map is used to describe the relationship between sentences. Topic sentences of a text can be got through searching the nodes which have most of edges. Using the semantic dictionary, sentence similarity is defined as its semantic similarity to address the problem of low word frequency similarity of short text. An internet public campus is chosen to take a test, 80.6% acceptability have been achieved.