The Research On Word Representation Theory And Its Application In Natural Language Process

Word embedding is the basis of natural language processing.It maps the words into the language that the computer can understand in form and meaning.This makes communication between people and computers simpler.The quality of word embedding has a great impact on natural language processing.In this paper,the theory of word embedding based on fuzzy sets is studied,and the application of word embedding in natural language processing is discussed.Firstly,we introduce the existing models of word embedding.We consider the influence of the different parameters such as the window size of the context,the dimension of word embedding and the training threshold of mininum word frequency.The cosine similarity,the analogism and the detection of synonym are used to evaluate the effect of different parameters on word embedding that trained by continuous-word-bag model.Secondly,a fuzzy query expansion method based on word representation and a fuzzy information retrieval model based on continuous bag of words model are established.We combine the theory of fuzzy mathematics and word embedding to calculate the degree of membership between retrieved text and query item.Finally,we put forward an integrated evaluation method of machine translation quality based on Z-number.Z-number can measure the reliability of fuzzy constraints.There are different word representation theories.People use the different word representation theories in machine translation.The method utilizes the characteristic of Z-number to evaluate the different machine translations.