框架元素语义核心词自动识别的研究.pdfVIP

 框架元素语义核心词自动识别研究 康旭珍，李双红，李茹 山西大学计算机与信息技术学院，山西太原，030006 E-mail: kangxuzhen@126.com 摘 要：本文基于汉语框架网，用框架核心依存图形式化的表示一个汉语句子，使得对句子能够进行深层 语义理解。为了得到框架核心依存图，需要提取框架元素的语义核心词。文中使用条件随机场模型和最大 熵模型来识别框架元素语义核心词，并分别对两个不同的模型所选的特征集进行了对比分析，且通过构造 不同的特征模板进行对比实验，选取了较优的特征模板和模型。结果表明，CRF模型具有较好的识别性能， 它在最优的特征模板下，对简单型和复合型短语类型框架元素语义核心词识别的平均正确率分别达到了 96.45%和95.17% 。 关键词：框架元素；框架核心依存图；条件随机场；最大熵模型 The Automatic Identification of the Semantic Core Words for Frame Elements KANG Xuzhen, LI Shuanghong, LI Ru School of Computer Information Technology, Shanxi University, Taiyuan 030006, China E-mail: kangxuzhen@126.com Abstract: In this paper, a Chinese sentence is formalized by the Frame Kernel Dependency Graph, which makes the deep semantic understanding of the sentence. To obtain the Frame Kernel Dependency Graph, it is necessary to extract the semantic core words of Frame Elements. This paper uses Conditional Random Fields model and Maximum entropy model to extract the semantic core words of Frame Elements, and feature sets with respect to these different two models were compared and different feature templates setting were conducted, selected the optimum template and model. Experimental results showed that CRF model has better performance to this research. With the optimal feature template, the average precision of experiment result achieved 96.45% and 95.17% for frame elements of simple and complex phrase type respectively. Keywords: Frame Elements ；Frame Kernel Dependency Graph ；Conditional Random Fields；Maximum entropy model 1 引 言 随着中文信息处理技术的不断发展，自然语言处理研究逐渐从句法处理转移到语义处 理和语用处理方面，许多研究工作尝试以新的角度来观察、发现新的语言问题。其中基于 框架语义学理论构建的框架语义网络知识库为自然语言处理开辟了一条新的道路。基于汉 语框架网，将一个汉语句子形式化表示成框架依存