Research On Sentiment Analysis For Product Review Based On Deep Learning

In the "Internet+" era,companies or businesses are keen to use Internet-based social platforms to collect consumer reviews of the goods or services they offer to improve the quality of their products or services.Automatic sentiment analysis based on product reviews is to achieve commercial value.The important means of this paper is to conduct in-depth research on the sentence-level sentiment analysis and aspect-level sentiment analysis of product reviews.The main contributions include:Aiming at the fixed coding problem of sentence-level sentiment classification algorithm based on Attention mechanism,this paper proposes a sentence-level sentiment classification model of hierarchical Long Short Term Memory and dynamic routing mechanism(LSTM-DRM),which uses the Attention mechanism to extract important word features in the first layer of LSTM;transfer feature information to the classification layer using a dynamic routing mechanism in the second LSTM layer dynamically.The experimental results show that compared with the single Attention mechanism,the LSTM-DRM model proposed in this paper can obtain higher classification accuracy and lower root mean square error in the three standard data sets of IMDB,Yelp2013 and Yelp2014.It can also verify the effectiveness of the dynamic routing mechanism in text sentiment classification tasks.The existing sentiment classification algorithm ignores the importance of user and product information in sentiment analysis in product reviews.Therefore,based on the sentensentiment classification model,this paper proposes LSMM-UPDRM that integrates user features and product features.The model respectively uses two independent LSTMDRM models to embed user and product features.The LSTM-UPDRM model first uses the Attention mechanism to extract specific words that users and commodities are explicit in the comments.Then,the dynamic routing mechanism is applied in the classification layer to extract the emotional features implicitly of users and commodities in the comments.Finally,the weighted loss function is used to characterize the importance of the two features to the emotional category,and the accuracy of the model is tested in the JD product review data set.Through comparative experiments shows,The LSTMUPDRM model in this paper can obtain 93.4% accuracy when used in the sentence-level sentiment classification task of product reviews,which is higher than other existing methods.Aiming at the problem that the current aspect-level sentiment classification algorithm does not consider the context information of the aspect words,this paper proposes and implements a dynamic routing model(ELMo-DRM)that integrates contextual word vectors to complete the attribute sentiment classification in product reviews.The ELMo-DRM model first uses the ELMo model to obtain the context word vector,and combines the aspect features into the text features through the corresponding aspect coding algorithm.The dynamic routing mechanism aggregates feature information with higher similarity to complete the aspect emotion classification.Through comparison experiments shows: the ELMo-DRM model proposed in this paper can obtain 90.2% better accuracy than the existing scheme in the product reviews aspect-level sentiment classification task.