Text Sentiment Analysis Based On Deep Learning

With the development of Internet technology,people can express their views directly on the Internet.For example,on shopping websites or other social networking sites,the content published includes products,as well as services and events.People show their emotional tendencies through the process of online reviews.These tendencies can provide important information for other people's subsequent purchase behaviors,and can also provide important help for competition between businesses.In this context,online review sentiment analysis technology has emerged,which can show people's emotional tendencies in online reviews.This paper first analyzes the key technologies of sentiment analysis of movie reviews,and then studies the specific application of the key technologies of sentiment analysis of reviews on the Internet.This paper mainly studies the models related to sentiment analysis from the following two aspects:First: Analyze movie review texts based on convolutional neural networks CNNs.The experiments and control experiments are CNNs-Word2vec-Part of speech,CNNsWord2 vec,CNNs-Rand-Part of speech,CNNs-Rand,SVM,SVM-Part of speech models.The experiments show that the CNNs-Word2vec-Part of speech group of experiments has the best emotion classification effect.Its accuracy is 0.891.Comparing the experimental results based on the CNN model with the results based on the SVM model,it is found that the sentiment analysis based on the CNN model has a better classification effect.After analyzing the results,it is found that the CNN model can extract deeper information of the text vector;At the same time,CNNs-Word2vec-Part of speech and CNNs-Word2 vec are compared,and CNNs-Rand-Part of speech and CNNs-Rand are compared.The results of the CNNs-Word2vec-Part of speech and CNNs-Rand-Part of speech models are better than those without the part of speech.This shows that considering part-of-speech features in the sentiment analysis model and removing some unused empty words significantly improves the accuracy of the results.Comparing CNNs-Word2 vec and CNNs-Rand models,it is found that the same CNNs model based on Word2vec's feature word vectors trained is significantly more accurate than randomly constructed word vectors.It is considered that the word2 vec tool can extract more original data features.Second: Explore the BERT model proposed by Google and find that the feature word vector output by BERT is re-entered into the CNN convolutional neural network for convolution operation,and the accuracy of emotion analysis is higher than that of using only the BERT model high.Considering that the CNNs convolutional neural network can better extract the local features of the text to be processed,and can also reduce the dimension of the emotional features,the sentiment analysis of the BERT model can improve the accuracy.