【作者】 钟茂生； 刘慧； 刘磊；

【Author】 ZHONG Mao sheng~(1,2)LIU Hui~1 LIU Lei~1 1.Department of Compute Science and Engineering,Shanghai Jiaotong University,ShangHai 200240:2.School of Information Engineering,East China Jiaotong University,NanChang,330013

【机构】 上海交通大学计算机科学与工程系； 华东交通大学信息工程学院；

【摘要】 词汇间语义关系的定量化研究是文本分类、信息检索和基于统计的语言模型等自然语言处理任务中一个重要的基础性工作。词汇间语义关系总体上分为等同关系、上下位关系、相关关系,现有的语义关系定量化工作主要集中于词汇间语义的等同关系(相似性)量化研究。本文研究了词汇间语义的相关关系定量化问题,分析了用协方差、共现频率和互信息等工具对词汇间语义相关关系进行定量化存在的缺陷,提出了量化词汇间语义相关关系的基本思路和新方法,并构造词汇相关关系二分图来求解和量化词汇间间接相关关系,该方法能够解决在统计语料中没有出现的词汇对的相关关系量化求解问题。实验结果表明,本文提出的方法比单纯用互信息来计算和量化词汇间语义相关关系更为可行。同时,对于一个特定词汇而言,本文的方法能够得到一个相关关系量化的相对合理的趋势性结果。更多还原

【Abstract】 the quantitative research of semantic relation between words is an important and basic subtask for some natural language processing task,such as Text Classification,Information Retrieval and Statistical-based Language Models.Generally,semantic relation between words concludes three types relations,namely, Synonymy Relation,Hyponymy Relation and Relevance Relation.The existing quantitative research of semantic relation between words mostly focus on how to quantity the Synonymy Relation(or similarity relation)between words.We have studied on the problem of Semantic Relevance Relation measurement of between words in this paper,and analyzed the shortcomings of these tools,such as Probability Covariance,Co-occurrence Frequency and Mutual Information,for the task of Semantic Relevance Relation measurement of between words.Then,we presented a novel approach to quantity the Semantic Relevance Relation between words,and constructed Bipartite Graph of Lexical Relevance Relation for measuring the indirect relevance relation between words. Moreover,our approach can resolve the problem of how to measure the Semantic Relevance Relation between words,which do not co-occur in the Corpus.The experiment results showed that our approach to measure the Semantic Relevance Relation between words is more feasible than Mutual Information tool.At the same time,for a special word,our approach can obtain a relative reasonable trend result on semantic relevance relation measurement between words.更多还原