Research On Short Text Emotional Tendency Analysis Based On Deep Learning

With the development of social media and e-commerce,more and more users are beginning to use social networking services to express their opinions on various topics and objects such as commodities,public figures,and news events.If we can extract valuable emotional information from these massive information,it will promote the development of product recommendation,public opinion monitoring,and public opinion research.The text sentiment orientation analysis aims to judge the emotional polarity of the text with emotional color.It is a core task of text sentiment analysis.It has very important research value and faces many challenges.The classification method based on deep learning has good adaptability,and the feature learning ability is relatively strong and has great competitiveness.However,the existing deep learning methods are relatively simple in the representation of text feature vectors,and the keywords in the text data are not effectively utilized.Therefore,in view of the existing problems,the main research work of this paper is as follows:(1)For the problem of text representation,a Multi-Granularity Fusion Convolution Neural Networks(MGF-CNN)text sentiment classification algorithm is proposed.Firstly,by combining the part of speech features,location features and word vector features,a text representation method for multi-granularity feature fusion is proposed.Feature extraction is then performed by Convolution Neural Networks(CNN).In order to extract more important text feature information,this paper uses different size convolution window and maximum pooling operation method.Finally,the algorithm was tested in the Chinese and English comment data sets.The experimental results show that the accuracy of the MGF-CNN model is significantly higher than that of the single-word vector deep learning algorithm.(2)In order to better extract the sequence characteristics of the sentence,more attention is paid to the keywords in the sentence.This paper proposes to combine the CNN and Bidirectional Long Short-Term Memory(BiLSTM),and add the attention feature to the Attention Mechanism to design the CBLSTM-Attention model.Firstly,using CNN to extract the local features of sentences,using BiLSTM to extract context sequence features,the features extracted by the two neural networks are combined to obtain a more comprehensive text semantic expression.Then,the merged features are given different weights to different words through the attention model based on the gating mechanism,so that the model pays more attention to the features related to the output results.The experiments prove that the CBLSTM-Attention model improves the accuracy of text classification on the four data sets of Chinese and English than other representative papers.