Research On Emotional Tendency Classification Method Of Microblog Comments Based On Deep Learning And Attention Mechanism

With the development of China entering a new era,all kinds of real-time media continue to develop and mature,and have become the main channel for the voice of the majority of Internet users.Microblog is the most mainstream and most widely accepted social platform at present,carrying the thoughts and views of many netizens on different events.At the same time,big data technology and natural language processing technology are also developing and improving.In this context,a good analysis of the emotional tendency of microblog users' comments is not only conducive to the platform to listen to users' voices widely,but also provide important reference for decision makers in the process of public opinion processing and controlling public opinion guidance.At present,most of the emotional tendency analysis of microblog comment text data divides the emotional tendency into positive emotion,neutral emotion and negative emotion,but both positive emotion and negative emotion are too generalized to be accurate to a certain emotion.On the other hand,the text data of microblog comments are expressed by netizens,and the format is often not standardized.At the same time,there are long comments in the microblog comment text data.To obtain the emotional tendency,we must closely combine the semantic information between contexts.In view of the above situation,the main innovations and research results of this paper are as follows:1.In order to better obtain the semantic features of microblog text information and integrate the text features into the emotional tendency classification model,this paper proposes an emotional tendency classification model based on BiLSTM,which preliminarily realizes the recognition of microblog emotional tendency at the sentence level.2.The BiLSTM model is used to process the text data of microblog comments.On this basis,combined with the attention mechanism to further improve the accuracy of emotion Tendency Classification,a BiLSTM attention model with five kinds of emotions is proposed.3.Because the format of microblog comment text data often disagrees,this paper proposes to expand the training set of microblog comment text data to improve the accuracy of emotional tendency classification of the model and better identify abbreviations,nicknames and so on.4.Due to the existence of long comments in the microblog comment text data,and the conventional method of calculating text similarity is difficult to capture the semantic relationship in the long comment text,in order to further improve the accuracy of emotional tendency classification,this paper innovatively proposes to apply the word shift distance WMD(word mover's distance)algorithm to integrate with the attention mechanism,and use the WMD algorithm to calculate the text similarity in the attention mechanism,combined with the two-way long-term and short-term-memory network,BiLSTMWMD-Attention affective tendency classification model is proposed,and the effectiveness of the model is verified by multiple groups of comparative experiments.Compared with BiLSTM-Attention,BiLSTM-WMD-Attention has great advantages in all evaluation criteria.The highest ACC value is 0.099,the highest pre value is 0.141,the highest rec value is 0.12,and the highest F1 value is 0.116.At the same time,experiments verify the respective advantages of BiLSTM-Attention model and BiLSTM-WMD-Attention model with five kinds of emotions in different microblog comment text data.