Research On Sentiment Analysis Of Comment Text Based On Deep Semantic Features

With the rapid development of the Internet,various types of online commentary have become more and more popular.The sentiment analysis of these commentary texts can provide assistance for personal consumption decision-making,business marketing strategy planning,and government public opinion detection.On the basis of summarizing the existing research and technology related to sentiment analysis,this paper uses deep learning technology to mine the deep semantic features implicit in context information,so as to conduct sentiment analysis on the review text.The main work is as follows:(1)Aiming at the shortcomings of traditional feature extraction methods and word vector usage,the paper combines with the advantages of bidirectional recurrent neural network which can better express context information and convolutional neural network for feature extraction.A joint textual sentiment analysis method based on Bidirectional Recurrent Convolutional Neural Network(BR-CNN)is proposed.Firstly,the PV-DM model is used to train the word vector with rich semantic information.Then the BRNN is used to learn the context information vector of the word,and the CNN model is constructed for deep semantic feature extraction.The comparative experiments on the public dataset show that the BR-CNN model has good classification performance and good generalization ability,and the accuracy rate is91.6%.(2)The combination of Attention Model(AM)can highlight the contribution of each word and the advantages of Bidirectional Long-Short Term Memory(BLSTM)in dealing with long-distance dependent information.Based on the BR-CNN model,AM is added,and BRNN is replaced with a better performance BLSTM model,and a new network model AC-BLSTM is constructed.Firstly,the Glove model is used to train the word vector with rich global statistical information.Then the first round of deep feature mining is performed by BLSTM,the context semantic information is fullyretained,and then the most relevant part is automatically found from the stored information by using AM.To give the largest feature weight,and finally use CNN for quadratic feature mining.On the public dataset,a comparison study between AC-BLSTM and BR-CNN showed that the accuracy rate increased by 2.3% to 93.9%.