基于文档相似度的双语文档排序学习  

Learning to Rank Bilingual Document Based on Document Similarity

作　　者：黄健[1] 

机构地区：[1]昆明理工大学智能信息处理重点实验室,昆明650500

出　　处：《计算机与数字工程》2017年第10期1986-1989,2017,共5页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61175068;61472168);云南省关键项目科学基金项目(编号:2013FA130);科技部科学技术创新人才项目(编号:2014HE001)资助

摘　　要：论文提出了一种双语文档排序学习模型。排序是信息检索系统中重要的一个环节,学者们已经对单语言环境下的文档排序学习问题进行了很多的研究,但是多语言语境下的文档排序问题并没有得到很大的重视。在双语文档集合中,文档是用两种不同语言写成的,所以我们并不能使用已有的单语言排序学习模型直接对双语言文档进行排序。在单语言排序学习模型的基础上,提出了一种基于双语文档相似度的双语文档排序学习方法。为了能够把源语言文档和目标语言文档关联起来,提出了一种基于双语word embedding的双语文档相似度计算方法。通过简单地把源语言查询翻译为目标语言,达到筛选目标语言文档的目的。实验结果表明,该方法在英汉和英越两种语境下都取得了较好的效果。The problem of learning to rank bilingual documents is addressed. Ranking is an essential part in information retrieval. Ranking documents in monolingual context using machine learning has been studied a lot,but learning to rank bilingual documents has not been investigated much yet. Bilingual documents are written in different languages,they can't be processed by using existing monolingual methods directly. In this paper a bilingual learning is proposed to rank model which utilizes monolingual model to give ranking score for documents in monolingual context as a base component. A word embedding approach is introduced to measure document similarity in bilingual context,through which a relationship between documents in both languages can be made. We simply translate the query to foreign language at a phrase level to filter foreign language documents. Experiments show that our model is effective in ranking bilingual documents in both English-Chinese context and English-Vietnamese context.

关 键 词：排序学习 信息检索 文档相似度 查询翻译 双语语境 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]