Abstract: Focused on the issue that computer cannot automatically carry out a pragmatic analysis of the deep meaning    of whole discourse at present, this paper designs the relevance inference algorithm based on fuzzy conceptual graph. In the algorithm, aiming at the specific dialog mode of Chinese pragmatic analysis, the discourses of speakers and the knowledge of cognitive context are expressed in fuzzy conceptual graph, and the relevance inference is conducted from computer science. The problem that computer automatically deduces the deep meaning of whole discourse is resolved. Through the experimental analysis, accuracy can reach 78%. In addition, the algorithm has been applied in analyzing public opinion and mining social network. After the preprocessed discourses of speaker by this relevance inference algorithm based on fuzzy conceptual graph, this algorithm can reduce the missed alarm rate and false alarm rate of Mutton and AdaBoost methods based on multi-features fusion and increase the accuracy of SBV polar transfer algorithm. The algorithm can deduce the deeper meaning of answerer??s discourse at specific dialog mode.

Key words:  pragmatic analysis, relevance inference, fuzzy conceptual graph, cognitive context

摘要： 针对目前计算机在自动语用分析中不能解析出整个话语深层含义的问题，设计了基于模糊概念图匹配的关联推理算法。该算法针对汉语语用分析中的特定对话模式，用模糊概念图表示说话人的话语和认知语境知识，并从计算机学科出发进行关联推理，解决了话语深层含义的语用分析问题。经过实验分析，该算法准确率达78%。该算法已应用到舆情分析和IRC聊天室社会网络挖掘中，采用该算法对大量会话文本预处理，有效降低了基于多特征融合的Mutton方法和AdaBoost方法的漏报率和误报率，提高SBV极性传递算法的准确率，有效推出了对话者文本的深层含义。

关键词: 语用分析, 关联推理, 模糊概念图, 认知语境