Research On Chinese Word Sense Disambiguation Method Based On Graph Model

As a basic research of natural language processing,word sense disambiguation(WSD)has important influence on upper layer applications,such as machine translation,information retrieval,text classification and sentiment analysis.The bottleneck of knowledge acquisition is an important factor that restricts the development of WSD technology.The multiple existing knowledge resources have their own advantages.If we can integrate a variety of Chinese and English knowledge resources to complement each other,and fully exploit the knowledge of disambiguation in these resources,it will help the development of WSD.From this point of view,this dissertation attempts to carry out relevant research.In the framework of graph model,a variety of Chinese and English knowledge resources are used for similarity computation,and simulated annealing algorithm is used to optimize the similarity values.The main work and contributions of this dissertation are listed in the following three aspects:(1)Aiming at the shortage of knowledge resources faced by traditional Chinese WSD methods,this dissertation proposes a method of WSD based on similarity computation of English words and graph model.The main idea is to convert the problem of Chinese WSD to the English field,and then use the more complete English knowledge resources to deal with WSD.For the two key problems involved in this method,i.e.word meaning mapping and similarity computation of English words,this dissertation proposes the method of mapping senses in BabelNet based on word embedding,and the method of word similarity based on word embedding and knowledge base.Experimental results show that this method can make full use of English knowledge resources and improve the effectiveness of Chinese WSD.(2)Aiming at the insufficient utilization of HowNet knowledge in the existing disambiguation methods,this dissertation proposes a method of WSD based on HowNet and graph model.This method uses the techniques of dependency parsing to acquire contextual knowledge,and to construct the dependency disambiguation graph.By the means of dependency parsing,this method processes the examples in HowNet which have a good ability of sense distinction,to construct the dependency disambiguation graph.Then it completes the process of disambiguation by combining dependency disambiguation graph and contextual disambiguation graph.Experimental results show that this method get the disambiguation accuracy of 0.468 on the dataset of SemEval-2007 task#5,which are better than others in the same type.(3)To further integrate various types of disambiguation knowledge resources,this dissertation proposes a method of Chinese WSD based on graph model.This method is the integration of aforementioned work to further improve the performance of WSD.A weight optimization algorithm based on simulated annealing is designed,by which the parameters of three similarity values are optimized.Then the disambiguation graph is constructed to process the task of WSD.Experimental results show that this method can improve the effectiveness of Chinese WSD,which obtains the disambiguation accuracy of 0.492 on the dataset of SemEval-2007 task#5.