Research On Automatic Construction Technology Of Chinese Verb Frame Base

Semantic knowledge base can almost be used in any aspect of Natural Language Processing, such as syntactic analysis, machine translation, word sense disambiguation, information extraction and so on. Verb is the core of a sentence, so the building of semantic knowledge base about verb is rather important. However, at present most semantic base construction methods are based on manual work. If we want to build a base which is applicable from the aspects of scale and updating speed, lots of manual work, material and financial resources are needed. This paper focuses on the application of verb semantic information in NLP, and proposes a method of automatic construction of Chinese verb frame base.Firstly, through analyzing a large number of Chinese sentences, this paper summarizes the characteristics of some specific cases'case marks, keywords and keyword positions. Secondly, we observed the nature of the phrases which are the same case of the same verb. In response to these characteristics, this paper divides the automatic building process into two parts, namely verb case relationship identification and semantic class acquisition. Correspondingly, this paper proposes an automatic verb case relationship identification method. This method extracts different features according to case relation and proposes the feature of the case relation mark, the kind of dependency relationship .Then a maximum entropy model is trained to classify the case relation. This paper also modifies the classification results by adopting a rule-based post-processing strategy. Besides, this paper proposes an automatic semantic class acquisition method of the necessary verb case relations. This method uses an algorithm based on available word similarity to acquire semantic class of the necessary verb case .And this paper improve the performance of the algorithm through the hierarchical processing strategy, and heuristic information.Experimental results show that the precision of case relationship identification reaches 90.70%, and the precision of semantic class acquisition reaches 98.00%.