The Research Of Sentiment Analysis Of Civil Aviation Microblog Text Based On Deep Learning

Today,Internet penetration is so high and the number of Internet users is so large,people prefer to express their opinions and get hot news in a fast and concise way more and more.As a new comprehensive platform,microblog can meet the needs of people to get news and express opinions.Sentiment analysis of a large number of microblog texts which containing emotional information can be used to effectively monitor public opinion and marketing strategy,etc.In this thesis,the deep learning method is used to conduct sentiment analysis on microblog containing civil aviation key words.There are two main components in this task: text vectorization representation and semantic information extraction.The former uses the neural network to learn the vectorized representation of semantic information from the massive text data;the latter uses the deep learning model to further extract high-level semantic features and syntactic features based on the vector representation of the text.Through the analysis and research of existing methods,the deficiencies of these two parts are improved.(1)Aiming at the problem that the emotional information contained in the existing word embedding models is not enough,the word embedding model of fused emoticons is given in this thesis.The model makes full use of the more explicit characteristics of the emoji emotional information in the microblog text,and correlates the emoji word embedding with the original word embedding to construct a more lexical word embedding model with emotional semantic information.And then the convolutional neural network is used to train the sentiment analysis model based on the word embedding model.The parameters of the word embedding model are determined by a large number of experiments,and compared with other word embedding models,the accuracy rate is improved.(2)Aiming at the problem that the deep learning model structure is not comprehensive enough to obtain the dependence and location characteristics between texts,a multi-dimensional self-attention mechanism combining location information is given in this thesis.The method extends the self-attention mechanism to multi-dimensional calculation,and integrates the forward and backward position features.The model can effectively extract the multi-angle high-level semantic features of the microblog text and obtain the internal relationship between the words in the text.Experiment on datasets containing civil aviation lyrics keywords and public dataset COAE2014 task 4.The results show that the sentiment analysis model based on the multi-dimensional self-attention mechanism of the fusion location information is better than the comparison models.