Research Of Text Processing Method And Application Based On Attention Mechanism And Word Sense Disambiguation

The explosive growth of network information has prompted the development of computer technology for automatically processing text.How to obtain the semantic information efficiently and accurately is an urgent problem to be solved.Considering that the understanding of a sentence by the human brain can be seen as a process of retrieval of conceptual information and integration of semantic information,so this thesis aims to study related computational models and get more prior knowledge and more critical information to accurately eliminate the ambiguity of words and extract richer semantic information.The main work and research contents are as follows:(1)Research on the feature extraction method based on statistical features and event attention mechanisms.While solving the text processing problem,the existing Attention based Bi-directional LSTM has the problems of increasing the calculation of the model and losing the semantic information of the text.We propose to calculate attention weights at the structured event level.At the same time,considering the existing deep learning model which can`t learn the statistical feature of the text,this thesis add the statistical features on the basis of attention weight calculation.Experiments show that Compared with the existing models,the semantic information contained in the event structure and the corresponding statistical features improve the quality of the text vector representation,reduce the computational complexity of the model and improve the accuracy of text classification.(2)Research on the Word Sense Disambiguation method based on the Dualchannel LDA topic model.The existing word sense disambiguation method based on the LDA topic model mostly uses the document topic as the main basis of disambiguation.Although the whole document is used as the training corpus,the meaning of the ambiguous word itself and its context information are neglected,which may cause the problems of data sparseness and lead to deviations in the meaning of the words.In response to the above problem,this thesis proposes a word sense disambiguation method based on the dual-channel LDA topic model.We introducing the external knowledge base of the ambiguous words(namely the WordNet Synset)as a channel input for the LDA topic model,at the same time,the adjacent words of ambiguous are extracted as the input of another channel of LDA topic model.Therefore,we make full use of external knowledge and improve the accuracy of disambiguation.(3)Research on the neural machine translation based on event Attention mechanism and statistical machine translation method incorporating word sense disambiguation.For the problem that the existing neural machine translation method is relatively inaccurate for long sentence translation,applying the event-based Attention mechanism to machine translation;considering that statistical machine translation is relatively effective in a specific field,the word sense disambiguation method based on the dual-channel LDA topic model is incorporated into the phrase-based statistical machine translation model.The experimental results on the datasets selected in this chapter show that the neural machine translation of the embedded event Attention mechanism and the statistical machine translation incorporating the word sense disambiguation enhance the translation effect,indicating that the improved text processing method has certain practicality value in the field of machine translation.