Research And Application Of The Semantic Analysis Based On Conceptual Graph

AI has been greatly developed in 21 st century. As one of the key subjects in AI, natural language comprehension attracts more and more attention among researchers and has been applied in practice widely. Semantic analysis, an important branch of natural language comprehension, has shown its enormous value in research. Based on conventional Chinese semantic analysis, the present work has designed a prototype model for Chinese semantic analysis system with weighted conceptual graph as knowledge representation and reasoning(KR) and Chinese semantic knowledge base as data support. In this paper, the framework, the functions of modules, and the accuracy of semantic analysis are investigated for the prototype model.In the prototype model for Chinese semantic analysis system, a first requirement is the determination of knowledge representation and reasoning. As a mature KR, conceptual graph has a great faculty for semantic representation and knowledge reasoning after years of research and development. Its linear representation is fit for storage and operation in computer. Based upon conceptual graph, the concept of weight is introduced to construct the weighted conceptual graph in the present work. Furthermore, with the weighted conceptual graph combined with the knowledge of semantic similarity computation, and the prototype model is designed.In the prototype model, Chinese word segmentation, syntactic analysis, generation of conceptual graph, and similarity computation are included. The Chinese word segmentation uses Cascaded Hidden Markov Model to segment Chinese words. Then, for the segmented sentences, dependency parsing is employed to fulfill the syntactic analysis and to mark syntactic relation among words. In the generation of conceptual graph, common syntactic relations and semantic relations are summarized, and the rules of conversion between grammatical relations and semantic relations are studied. Using semantic database as data support, the rules of conversion are specified, based on which grammatical relations are converted to semantic relations. Thereby, Chinese sentence can be expressed in the form of conceptual graph. Finally, algorithm corresponding to similarity computation is designed and coded to accomplish similarity matching for each part of the conceptual graph. Thus, semantic analysis is realized.On the basis of these detailed parts, the present work has constructed a prototype model for Chinese semantic analysis system with weighted conceptual graph as knowledge representation and reasoning. This system is capable of segmentation, parsing and similarity computation for Chinese text. In debugging and comparison with experimental data, the matching accuracy reaches as high as 50%, and the success probability of syntactic recognition and transfer reaches 65%. For polysemy and synonyms in Chinese language, this system realizes an accuracy of 60%, higher than conventional matching method.