Research On Emotional Tendency Analysis Based On Film And Television Reviews

With the vigorous development of Internet e-commerce,a large number of user reviews data have emerged on the Internet.These reviews contain a lot of valuable information and become an important factor that affects business and user decisions.More and more users will see the merits and demerits of the product by browsing the reviews.By analyzing the comments,the business can better understand the needs of the users and find their own shortcomings and advantages.Network reviews have become an important source of collecting users' opinions and improving the quality of products.In this context,feature mining and emotional analysis emerge as the times require.But network reviews are random,and the quality is uneven,and the data sparsity is strong.The traditional algorithm has some shortcomings in the accuracy of feature extraction.But the traditional sentiment analysis using quantitative statistical review analysis concluded in commendatory and derogatory term,it is difficult to obtain the true emotional comments.The research topic of this paper is the research on feature extraction and emotional tendency analysis.The main purpose is to improve the feature selection algorithm,and combine the current popular learning technology and neural network model to analyze the large-scale text data in the network,and dig out the hot topics and emotional tendencies of users in the reviews.This paper mainly studies from the following aspects.(1)This paper puts forward based on mutual information algorithm,by introducing the relative frequency factor and combined with the feature weight way for the appropriate improvement of the mutual information algorithm selection in the lack of text features,the optimization of feature selection process,and combined with frequency and inverse document frequency algorithm for feature candidate feature matrix the dimension reduction process,so as to improve the efficiency and accuracy of feature extraction.Finally,it is proved by experiments that this method can effectively improve the problem of low accuracy of mutual information algorithm when dealing with massive and sparse data.The results of the experiment are compared and analyzed,which shows that the improved algorithm has a remarkable improvement effect.(2)This paper presents a deep learning WordtoVec method and neural network method LSTM length memory recurrent neural network algorithm combining the above vector method based on deep learning of the results through the trained neural network model to complete the automatic process of emotional analysis of the content of the data,so as to improve the effect of emotion the orientation analysis,so as to improve the accuracy of the product feature extraction and sentiment analysis Chinese text mining,and finally get the attention and emotional tendency of users from huge comments.Experiments show that better results of emotional analysis can be obtained by using this method.