Word Sense Disambiguation Technology Research Based On Hownet And Bayesian Model

Word sense disambiguation is a hot and difficult problem in the natural language processing, and it has an extremely important theoretical and practical significance to these aspect research such as the machine translation, the information retrieval, the syntax analysis and the text classification and so on. Because that the semantic knowledge acquisition bottleneck factor, the quality of word sense disambiguation knowledge base and the fit and unfit quality problem to word sense disambiguation model, the disambiguation accuracy has not been ideal too at this present stage. Therefore, how to further enhance the treatment effect of word sense disambiguation, has been the power and the goal for these scientific researchers in the word sense disambiguation field.Firstly, this thesis outlines the main methods of word sense disambiguation, and the main techniques of word sense disambiguation is reviewed. Next, it introduces these related concepts of word sense disambiguation, and these semantic classification systems such as<Tongyici Cilin>,<Semantic Knowledge-base of Contemporary Chinese>, HowNet and so on. Then, it discusses these related word sense disambiguation models including the decision tree and the decision table, neural network prediction model, maximum entropy, hidden Markov model, and Bayesian model. On this basis, it explaines the word sense disambiguation technology based on HowNet and Bayesian model. The content includes dependency grammar analysis, the architecture of word sense disambiguation based on HowNet and Bayesian model, the process of word sense discrimination based on HowNet, the Bayesian improved method, Bayesian inference based on HowNet and dependency grammar analysis, technological process of word sense disambiguation based on HowNet and Bayesian model, algorithm of word sense disambiguation based on HowNet and Bayesian model, and the model training and disambiguation case. Finally, the experiment test and the performance analysis of these Word sense disambiguation models are carried on, including experimental corpus, test description, test results, comparison and analysis. The experimental result includes that the neural network forecast model, the hidden Markov model, the Bayesian model and based on HowNet and the Bayesian model. The contrast and the analysis includes these experiment check analysises between the neural network model and the Bayesian model, the hidden Markov model and the Bayesian model, the Bayesian model and based on HowNet and the Bayesian model. The experiment indicates that the word sense disambiguation technology based on HowNet and Bayesian model by the author studied, has the quite prominent superiority in several kinds of word sense disambiguation model.