Research On Text Sentiment Analysis For Social Media Comments

With the advent of the Internet era,a large number of users have also flooded into social media,and users have generated a large amount of comment data in the process of using.These text data often contain users’ attitudes towards specific objects.Mining and analyzing these comment texts will bring great help to the enterprise’s product analysis and image building,the government’s public opinion control and policy implementation,and the consumer choice of ordinary users.Text sentiment analysis can quickly and efficiently judge the emotional tendency of these comments with subjective emotional information,thus helping users obtain core information.Therefore,sentiment analysis for social media has practical research significance.This paper mainly studies the classification of comment sentiment on social platforms,and constructs a deep learning model to realize the task of text sentiment analysis.The main work of this paper is as follows:(1)Aiming at the problem that the existing affective analysis model can not focus on the construction of syntactic dependency tree,which makes it difficult to make full use of syntactic information,this paper proposes a multi-fusion simplifying graph convolution network(MFSGC)model that integrates part of speech and external knowledge for aspect-level sentiment analysis.Use the bidirectional short-term and short-term memory network to obtain the semantic information of sentences,use external knowledge to strengthen the role of emotional words in sentences,and filter invalid part of speech combinations to prune the syntactic dependency tree,propose a multi-fusion matrix algorithm to combine the two to obtain syntactic information,and use the simplified graph convolution network to fuse the semantic information and syntactic information,so as to reduce the computational load and make the model more efficient.Experimental results on five open data sets show that the proposed improved method is effective and can significantly improve the performance of the model.(2)In view of the lack of syntactic information caused by the use of a single attention mechanism in existing models,the limited semantic information obtained by text coding and the lack of data set of Chinese drama review sentiment analysis,this paper proposes a simplifying graph convolution network based on Bert and interactive attention mechanism(IASGC-Bert)is proposed for aspect-level sentiment analysis.Climb the Douban Chinese drama review area to get the drama review data set.In order to enhance the emotional characteristics of the text,build a replacement dictionary,replace the emoticons,network words and professional terms in the review text,analyze the key words of the drama review and use the sorted data set in the subsequent aspect-level emotional analysis task.The Bert pre-training model is used to encode and represent the text,obtain the word vector containing rich semantic information,and use the emotion dictionary to strengthen the feature information of emotion words.In addition,the interactive attention mechanism is used to interactively learn the context and aspect words,fully obtain the syntactic and semantic information in the text,thus improving the performance of the model.The experimental results on the Chinese drama review data set and five English data sets show that the proposed improved method is effective and can significantly improve the performance of the model.(3)In order to visualize aspect level sentiment analysis,a text sentiment analysis system for social media is designed.The system calls the trained classification model to realize the Chinese and English sentiment analysis function.Through the word cloud chart,bar chart and pie chart,it displays and analyzes the drama review data set from multiple angles,and designs the overall page layout.Users can click different buttons to get different results feedback.