文献检索与基于影响的摘要系统设计与实现

Literature Retrieval System Implementation and Impact-based Summarization

构建了一种新型文献检索系统,能够摘要一篇文献中引起读者研究工作关注的那些内容,并返回读者对这些内容的评论,从而帮助用户快速了解该文献的学术价值及不足之处等重要信息。利用文献间的引用关系从其他文献中找到指向一篇文献的评论上下文,借鉴查询-检索模式,将评论转化为一元语言模型所生成的查询,并将原文献划分为句子所构成的文档集,基于KL-divergence检索模型找到原文献中与评论对应的句子。选取得分最高的若干句子构成体现原文献对外影响的摘要。系统基于北京大学研制的智能搜索引擎平台Platform for Applying,Researching And Developing Intelligent Search Engine(PARADISE),具有快速构建可扩展好的优点。

A new literature retrieval system is built to return the summary of a paper based on its literature impact. It can also return the comments given by other papers. The summary and the comments help readers quickly understand both the value of the paper and its inadequacy,which may not be found in the paper＇s abstract. The comments found in citing papers are virtualized as ＂query＂ generated by uni-gram language model,and the sentences of original paper are treated as a set of ＂document＂. Using KL-divergence scoring approach to find the similarity between the query and the documents,the top-scored impact sentences of the original paper are selected and returned as summary. The corresponding comments in the citing papers are also returned. The System is based on the Platform for Applying,Researching And Developing Intelligent Search Engine （PARADISE） developed by Peking University. It has the advantage of quick-to-start and good scalability.