Research On Frame Semantic Structure Analysis Technology For Chinese Sentences

Semantic analysis is most significant and difficult problem in natural language process fields. One of major goals that scholars both at home and abroad are all concerned about is how to realize effective, deep and automatic semantic analysis of sentences. Frame semantic structure of Chinese sentences is based on Frame Semantics and represents formally semantic structure of sentences with the aid of Chinese FrameNet of Shanxi University. This paper develop the core technology research of Chinese sentence frame semantic structure model, target words identification, frame disambiguation and frame semantic role labeling, meanwhile conduct the application research of tourism question-answering system based on the Chinese frames semantic analysis. The main research results as follow:(1) According to semantics struture of Chinese sentences, this thesis systematically analyzes frame semantic struture of Chinese sentences, puts forward frame semantic dependency graph models, including single frame semantic dependency graph, complete frame semantic dependency graph and core frame semantic dependency graph, and provides a novel way to represent semantic struture of Chinese sentences.(2) According to target words identification, the paper proposes the methods of unknown target words based on similarity computation and maximum entropy model. Because of full consideration words senses, dependency feature and context feature, they solve unknown target words identification and realize automatic extending to lexical units.(3) According to frame disambiguation, this thesis proposes a method based on T-CRF for frame disambiguation. It promotes the performance of frame disambiguation by adding long dependency relations in dependency feature. In addition, it compares with the frame disambiguation methods based on SVM and Maximum Entropy, which verifies the effectiveness of the frame disambiguation method based on T-CRF.(4) According to frame semantic role labeling, after summarizing and contrasting currently popular algorithms, we bring forward frame semantic role labeling based on T-CRF, which promotes labeling precision by adding dependency feature. Besides, based on frame semantic role labeling, we conduct the similarity computation from the point of frame semantics and put forward the semantic similarity computation of sentences based on multiple frames and significantion. The results of similarity computation based on frame semantic role labeling verify effectiveness of frame semantic role labeling for semantic similarity computation of sentences.(5) According to application research of semantic resource and semantic analysis methods of Chinese FrameNet, this paper designs and implements tourism question-answering system that orients tourism fields of Shanxi Province. The system takes Wutai Mountain for example, labels full-text frame semantic role to each introduction of scenic spots. The system includes question input, question analysis and answer extraction, which verifies the effectiveness of tourism question-answering system based on Chinese frame semantic analysis.The fruits of this paper further enrich theory and method for Frame semantic structure analysis research of Chinese sentences, provide a novel way to realize deep understanding of the semantic for Chinese sentences and provide a new technical support for more application systems based on semantic analysis in natural language processing fields.