中文句子语义概念图自动构建方法及应用研究

项目名称： 中文句子语义概念图自动构建方法及应用研究

项目编号： No.61462027

项目类型： 地区科学基金项目

立项/批准年度： 2015

项目学科： 计算机科学学科

项目作者： 钟茂生

作者单位： 江西师范大学

项目金额： 45万元

中文摘要： 概念图是被证明优于经典的、扩展的人工智能知识表示方法，构建自然语言语义概念图，将为自然语言处理模式探索新的道路。但是如何自动解析自然语言语义和自动构建概念图，还没有切实可行的方案。课题紧密结合汉语词语概念耦合的特点，借鉴逻辑学揭示概念内涵的思想，以内涵语义概念图为知识表示框架，以概念内涵结构为基本概念结构，用实体抽取和基于模板的概念关系抽取方法来获取句子内部概念结构，以句子模板概念图为骨架，用定义的概念图基本操作和自顶向下逐步求精策略，递归的将概念结构连接扩展并构建生成句子语义概念图，同时进一步研究基于示例学习的概念图自动构建方法，最后将概念图构建策略应用于科技成果奖励重复申报的文本复制检测。课题的研究将为汉语语义概念图自动构建提供新的思路，准确地构建句子语义概念图，对于改进信息检索、机器翻译、文本复制检测等任务的性能具有重要意义。

中文关键词： 自然语言处理；语义表示；概念语义分析；概念图；文本复制检测

英文摘要： Conceptual Graph is an extended artificial intelligence knowledge representation method, which is proven to be superior to the classical knowledge representation method. If there is a method to construct automatically semantic conceptual graph for the natural language, which will explore a new way for the natural language processing models. Though there is no feasible solution on how automatically to parse the semantics of the natural language and to construct concept graph up to now. In the project, by connecting closely with the characteristics that the concept of Chinese words is coupling, learning from the ideological that logic reveals the connotations of concept, using the connotation semantic conceptual graph for the knowledge representation framework and using the concept connotation structure<E, A, V> as basic conceptual structure, we present a method of constructing automatically the conceptual graph for the Chinese sentence, which include two steps, that is, the first step of building all <E,A,V> basic conceptual structures of this sentence based on these components, such as the 'Entity', 'Attribute' and 'Value', which are extracted by using the entity extraction and the conceptual relation extraction method based on the template, and the second step of connecting and extending recursively these <E, A, V> basic conceptual structures with the sentence template conceptual graph skeleton by applying some defined operations on concept graph and using the top-down stepwise refinement strategy. At the same time, we will further study automatic construction method of concept graph based on the way of learning from cases in the project. Lastly, we will apply the construction strategy of conceptual graph to the task of the text copy detection for science and technology achievement awards application documents. The researching of the project will provide a new way to automatically construct the semantic conceptual graph of Chinese sentences, and has influences on improving the performance of information retrieval, machine translation, text copy detection and so on by building accurately the sentence semantic conceptual graph.

英文关键词： Natural Language Processing;Semantic Representation;Conceptual Semantic Analysis;Conceptual Graph;Text Copy Detection