基于滑动窗口的动态摘要算法  

Query-biased Summarization Based on Slide Windows

作　　者：蔡建山[1] 迟呈英[1] 战学刚[1] 王丫[1] 

机构地区：[1]鞍山科技大学计算机科学与技术学院,鞍山114044

出　　处：《计算机工程》2007年第6期213-215,238,共4页Computer Engineering

基　　金：辽宁省自然科学基金资助项目(2004D110)

摘　　要：动态摘要是根据查询检索词从文章中动态提取的摘要。用户仅仅浏览动态摘要之后就能了解文章中与查询相关的部分,进而判断是否值得详细阅读整篇文章。该文根据搜索引擎对摘要速度和质量的要求,提出了一种使用滑动窗口抽取片断的算法,接着构造了摘要评测模型,使用同一个测试集对新动态摘要算法和Google、百度作对比实验。结果证明使用新方法生成的摘要能够言简意赅地概括文章的相关内容,在摘要指标的分项测试中取得了和Google基本相同的效果,但明显要比百度好,综合评价分别提高了5%和11%。A dynamic summary is based on the key words extracted from a text. By browsing the summary, a user can get knowledge of whether the document is relevant to his interest, and decide whether to read the whole text. According to the requirement of search engines on the speed and quality of a summarization system. This paper proposes a snippet extraction algorithm using slide windows. It also presents a performance evaluation of dynamic summary. Using the same document set, it performs an experiment contrast with Google and Baidu. Results show that the summary generated by this algorithm concisely outlines the content of a text, and compared with Google and Baidu, shows a better average performance, 5% than Google and 10% than Baidu. According to the respective evaluation of this experiment, this algorithm is similar to Google and obvious better than Baidu.

关 键 词：动态摘要 文本摘要 片断抽取 滑动窗口 

分 类 号：TP182[自动化与计算机技术—控制理论与控制工程]