Chinese Word Semantic Relation Classification Based On Neural Network Model

With the development of machine learning and deep learning technology in re-cent years,fields like computer vision,statistical natural language processing(NLP)and speech recognition have made a great progress.In the field of NLP,semantic under-standing has always been one of the important domain of research,and the classification of semantic relations between words is a foundation part of semantic understanding.Word semantic relation classification is a challenging task in NLP field,in many of its practical applications,we need recognize different kinds of semantic relations clearly.For example,we need recognize words with contrastive meanings in sentiment classifi-cation;in query expansion for information retrieval,we should recognize the synonym and hyponym of the target words.Among the research methods of semantic relation classification,there is a semantic dictionary matching method,which searches for semantic relationships between words matched by searching for semantic dictionaries.As for high-quality Chinese semantic dictionaries,there are Tongyici Cilin,HowNet and so on.There is no need to build a model in semantic relations classification by use of semantic dictionaries.Moreover,this method is simple,efficient and easy to understand,but a high-quality semantic dictionary is not easy to obtain for it needs to consume a lot of manpower and material resources and be regularly updated and maintained,it cannot recognize unregistered words as well.Method based on machine learning overcomes the defect of semantic dictionary.At present,there are many mature machine learning algorithms,such as LR(Logistic Regression),Markov model and so on.The general steps of such methods are converting words into vectors and then use machine learning algorithms to build models.Although it overcomes the defect of semantic dictionary to some extent,the accuracy of this method is not as high as the semantic dictionary method.In recent years,deep learning technology has been greatly developed.By establishing neural networks model or deep neural networks model,many difficult problems have been solved.Therefore,the neural networks model was used to research the semantic relations of words hereof.Using neural networks to build a model,it need to be cleared at first that the input and output of the network and network over-fitting problems need to be prevented as well.The input of network was designed here so that the input layer could contain all the semantic information.Meanwhile,some algorithms to prevent network over-fitting problems and improve the network performance was used,such as,dropout,regularization,exponential decay learning rate and so on.