NN型复合结构的语义关系识别及相似度计算  

RECOGNIZING SEMANTIC RELATIONS AND COMPUTING SIMILARITY BETWEEN CHINESE NOUN-NOUN COMPOUNDS

作　　者：张韬文[1] 陆汝占[1] 

机构地区：[1]上海交通大学计算机科学与工程系,上海200240

出　　处：《计算机应用与软件》2011年第3期5-7,12,共4页Computer Applications and Software

基　　金：国家自然科学基金面上项目(60873135)

摘　　要：目前信息检索的正确率不太高,原因之一是用现有的检索模型难以表示完整的用户查询意图,而用户在查询中大量使用了复合结构。通过实例探索了汉语NN型复合结构基于概念图的语义关系标引,发现复合结构的关联语义关系可以通过子成分的上下文求解。这些上下文通过网络进行识别抽取,并借助《同义词词林》进行泛化以解决数据稀疏性问题。复合结构内部的语义关系用向量来表示,向量的每一维代表了能表示复合结构语义关系的一个上下文。实验表明,提出的方法取得了较好的结果。The information retrieving model used today can＇t represent users＇ entire query intensions.This is one of the causes that the precision of retrieval is relatively low at present.Compounds are used a lot by users in query.In this paper we probed with instance the conceptual graph-based semantic relations indexing of Chinese noun-noun compounds and found that the associated semantic relations of compound can be resolved through the context of sub-component.These contexts are recognized and extracted through internet and generalized by means of＂Synonyms Dictionary＂to solve the problem of data sparsity.The semantic relations inside the compound are represented by the vector.Each dimension of the vector represents a context which can express semantic relation of the compound.Experiment shows that the method in this paper achieves competitive results.

关 键 词：复合结构 语义关系 语义相似度 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]