【作者】 李斌；

【导师】 蒋少华；

【作者基本信息】 湖南师范大学 ， 软件工程， 2019， 硕士

【摘要】 随着国际互联网的普及和WEB2.0的兴起,信息传播的速度和途径都发生了巨大的变化,在Web Of Science、知网、维普等科研文献数据库建立之后,每个科研领域都出现了文献种类繁多、数量激增的现象。科研人员使用关键词检索文献时,检索结果中总会有一些关键词匹配但内容严重不符的其他主题文献混淆在内,仍需花费大量精力从检索列表中“择出”与搜索关键词真正相匹配的目标文献,单纯从关键词上已难以甄别文献是否聚焦相同主题。为了帮助科研人员在拥有海量文献的科研文献数据库中找到与自己感兴趣文献具有相似主题的文献集,本文提出了一种基于语义扩展的综合三级文本相似度计算方法,该方法可以在文本的三种不同层次上挖掘两段文本之间潜在的语义相似关系,进而为用户感兴趣的文献筛选具有相似主题的文献集。为了在相似主题文献集中找到质量较高的科研文献,本文基于相似主题文献集内文献的作者、作者单位、下载量、引用量等信息提出了一个基于相似主题的文献质量推荐度模型,并通过仿真数据演示了该模型可以在相似主题文献集上把用户感兴趣且质量比较高的文献推荐给用户。在上述综合三级文本相似度和文献质量推荐度模型的基础上,本文设计并实现了一套相似主题科研文献自动推荐原型系统,该系统以科研人员提供的感兴趣文献为基础,到知网上爬取并筛选与该文献具有相似主题的文献集,然后在该相似主题文献集上通过计算文献质量推荐度,把科研人员感兴趣且质量较高的文献推荐给他们。该系统不仅能够根据用户感兴趣的文献进行相似主题文献推荐还可以对用户上传的文本提取关键词和摘要信息,后续的工作中还可加入用户画像技术,进一步增强系统个性化文献推荐的能力,更好地为科研人员服务。更多还原

【Abstract】 With the popularization of the Internet and WEB2.0,especially,after the establishment of CNKI,Web of Science,VIP database and other academic paper library,there has been an explosion in the number of academic papers in every research field.When researchers retrieve academic paper via keywords,they will always be confusion in the retrieval results,because it will always mix some unmatched retrieval results of other subject papers.They may still have to takes a lot of effort to "pick out" the target papers from the retrieval results that really matches the search keyword.After all,it is difficult to identify whether the papers focus on the same subject or not simply from the individual keywordsTo help researchers find out the papers collections with similar subject from millions of academic papers,a Comprehensive Three-level Text Similarity(CTTS)calculation method based on semantic extension is proposed,the CTTS can dig out the potential semantic similarity between two paragraphs at three different semantic levels,and then select the papers collections with similar subject.However,the significance of the selected papers is not the same.Users may prefer to see papers published in authoritative journals by authoritative people in the research field rather than those published in obscure journals by those authors without any influence.So,an academic paper recommendation model is proposed which is calculated according to author,affiliation,download number and citation number of the paper.The simulation data show that this model can be used to measure the referential value of academic papers,which is called as recommendation degree.When researchers upload a paper,the system will automatically crawls papers in CNKI according to the paper provided by users,and produces a papers collection with similar topics.After that,the system will automatically calculate the recommendation degree of each academic paper for recommend high-quality papers to researchers.The simulation data demonstrate that the model achieves the expected goal.In addition,the system can not only recommend similar subject papers according to the paper provided by users,but also extract keywords and generate abstracts according to the papers provided by users.In the followup work,user profile technology can be added to further enhance the ability of the academic paper recommendation system.更多还原