Sentiment Analysis Of User Reviews Based On Deep Learning

The rapid development of the Internet has made more and more users accustomed to sharing their own opinions and reviews on the Internet.Because of the sentiment information within the massive Internet user reviews can create great commercial and social value,sentiment analysis has long been a hot research topic in the field of information science and technology.The existing sentiment analysis model based on deep learning overcomes the shortcomings of traditional machine learning methods that rely on hand crafted features.However,the existing models cannot detect the importance of each part in a user review,and treat each local feature equally.This fact limits the performance of sentiment classification tasks.The word embedding technique maps words into low-dimensional vector space,which forms the inputs of deep learning text models.However,the current word vector learning models only use the semantic and syntactic information of the context words,but cannot obtain the sentiment information and domain relevance of words.This will cause some problems.First,some words with opposite sentiment polarity are too close in the word vector space.Second,some words that express different sentiment in different domains cannot be distinguished.Therefore,traditional word vectors are not ideal for direct application to sentiment classification tasks.In response to the above problems,this thesis has carried out the following research:To solve the problem that the existing neural network model cannot detect the importance of each word in reviews,a Convolutional Neural Network(CNN)sentiment classification model combining attention mechanism is proposed.The model first uses the convolutional neural network to extract the local features of the comments,and uses the Long Short-Term Memory(LSTM)network to extract the sequence features of the whole comment.By comparing the similarities between the two networks,the attention weights are calculated and the weighted sum of local features is used as the final representation for the sentiment classifier.After a large amount of data training,the attention mechanism can judge the importance of different words in a review,so that the model can “pay attention” to the most important local features extracted by CNN.The experimental results show that the proposed model can improve the performance of sentiment classification tasks compared with the traditional deep neural network approaches.Aiming at the fact that the current word vector model cannot obtain the sentiment information and domain information of a word at the same time,a Cross-Domain Sentiment Aware Word Embedding(CDSAWE)model for the sentiment analysis task is proposed.The model can extract sentiment information in the word by supervised learning.In addition,the model generates cross-domain word vectors for words by calculating domain correlations.These word vectors can capture the semantic change in domain adaptation.Through experimental verification,the proposed model can generate task-specific word vectors,and has better performance than several popular word vector models.