Research On Text Sentiment Analysis For Bullet Screen

With the rise of the short video industry,bullet screen has become a feature of players,and online video and e-commerce live broadcast have become the media of bullet screen.While watching the video,users can post comments on the current plot of the video through the emerging communication method of barrage,and have instant interaction with other viewers.Using text mining technology to emotional barrage data analysis to decision making for the user and program optimization and hot issues to control play a supplementary role.However,with the problems of open topic,fragmentation of content,diversity of evaluation objects and incomplete grammar structure,text analysis of barrage data becomes more and more complex.There is a certain research space on how to analyze the emotion of this special comment text.Emoticons and emojis Emotional symbols and emoticons,as common expressions of emotion in video barrage,are often ignored by the existing research institutes of barrage text sentiment analysis.In this paper,emoticons and emojis are selected to construct emoticons space,and the classification method of emoticons in barrage text is optimized by using the dual-attention mechanism.Therefore,the ES-MACNN model is constructed to analyze the sentiment tendency of the video barrage text.First,judge the importance of emotional symbols through the attention mechanism,then combine the content attention weight matrix with the emotional symbol attention weight matrix to get the semantic expression of the final text vector.Finally,use the convolutional neural network for feature extraction and analysis of emotional tendency of barrage text.In the experiment,the classification accuracy of the model is compared by comparing the data set with or without emotion symbols.On the data set without emotion symbols,the results of each model are not very different,but on the data set with emotion symbols,the method in this paper passes the introduction of emotion symbol space further improves the accuracy of emotion classification of barrage text,and the effect is significant.Aiming at the written features of the barrage text,which is severely spoken and hardly has complete contextual information,the multi-head attention based convolutional neural network model MH-ACNN was constructed to model and analyze the task of barrage text emotion multi-classification.Firstly,by encoding the position of words,the model can use the order information of the input sequence.Secondly,the multi-head attention mechanism is used to obtain semantic expressions in different subspaces,effectively capture the internal correlation of words,enhance the dependency relationship between words,and highlight the emotional weight of key emotional words.Finally,by combining the multi-head attention mechanism with a convolutional neural network,the seven emotion categories of the barrage text are modeled and analyzed.In the experiment,compared with the four emotion classification models,the MH-ACNN model has the best classification effect on F1.Based on this,according to the emotional tendency and emotional expression of the video barrage,the text of the video barrage is visualized and analyzed from multiple angles by combining the time-volume variation diagram and the emotion distribution diagram,so as to provide certain data support for video production and hot event control.