Emotional Analysis Based On Deep Learning And Personalized Recommendation Research

With the vigorous development of mobile Internet platform,more and more people tend to make objective comments on existing consumption,express their views,in this context,produced a large number of text comment data,which contains the real feelings of consumers on the product data.Mining the emotional experience contained in these text data and personalized recommendation analysis not only brings great convenience to consumers' quick choice,but also provides an effective reference for businesses to further develop their future development strategies.In the current study of emotional analysis,it is often tended to pay attention to the characteristics of the field of text,while ignoring the characteristics of the text itself,especially for the study of the combination of long and short text is lacking.At the same time,because of the diversity of the results of emotion analysis,there are great inadequate in the research of introducing personalized recommendation into the field of emotion analysis.From this point of view,this paper mainly carries on two aspects of research.On the one hand,the emotional analysis based on deep learning is carried out on the comments based on the combination of long and short text,and the optimal model under the characteristics of the material is selected by comparative experiment;On the other hand,on the basis of emotion analysis,the personalized recommendation research is carried out by using keyword technology and the similarity of Word Mover's Distance.Specifically,using reptile technology to obtain a combination of long and short text of hotel commentary materials,corpus cleaning,text word segmentation,remove thedeactivation word,convert text into word vector,and corpus tokenization were carried out,on the basis of which,based on the traditional RNN model,RNN variant(LSTM,GRU)model and Bi-RNN(Bi-LSTM,Bi-GRU)modelof emotional analysis,the results show that.The Bi-GRU model had the best emotional analysis under the characteristics of the corpus,and the accuracy of the test set reached 97.64%.On the basis of emotional analysis,Text Rank technology was used to extract the characteristics of positive and negative customer comments,and calculated the similarity of the customer's attention characteristics and the Word Mover's Distance of the positive comment text,use this as a basis for the different customer needs to be recommended,the experimental results showed that the top three recommended texts and customer demand were similar to more than 99%.The main contribution of this paper is to build 5 different RNN models for the comments that combined with long and short text,and to determine the optimal model for the characteristics of the material by comparison experiment.At the same time,on the basis of emotional analysis,a personalized recommendation study is carried out,which makes the results of text emotion analysis more valuable.