Fine-grained Sentiment Analysis Of E-commerce Platform User Reviews Based On Deep Learning

With the development of the Internet,a large number of e-commerce platforms have emerged.Fine-grained sentiment analysis of user review texts of e-commerce platforms can not only provide consumers with purchasing opinions,but also provide merchants with the advantages and disadvantages of goods.The results of granular sentiment analysis improve their products and improve the competitiveness of the company.At the same time,e-commerce platforms can apply the results of fine-grained sentiment analysis to areas such as personalized recommendations,intelligent search,and more.Since the method of deep learning needs to be applied to word vector,this thesis first improves the existing word vector,and then designs the two aspects of identifying the evaluation aspect from the user comment text and identifying the emotional polarity of the specific evaluation.Neural network model.The main work of this thesis is as follows:First of all,the current Chinese word vector often neglects the relationship between words and context,and can not solve the polysemy problem.Based on the existing word vector,this thesis applies the existing word vector based on long-term and short-term memory networks.Improvement,through the comparison experiment,the network structure of the best word vector classification effect is determined,and the classification accuracy of the new word vector obtained on various data sets is improved.By comparing and analyzing the advantages and disadvantages of existing word vectors and word vectors,this thesis designs a fusion model of word vectors and word vectors to obtain new word vectors.Experiments show that the classification effect of new word vectors is improved in word vectors.There are further improvements on the basis.Secondly,for the current model,it is impossible to identify the evaluation questions contained in the user’s comment text at one time.According to the current single label classification method,the single label loss function is improved to obtain the multi-label classification model.Experiments show that the improvement is obtained.The multi-label classification model solves the problem of identifying multiple evaluations from the user’s comment text at one time.By comparing experiments,the classification efficiency of the model is greatly improved under the premise of ensuring the classification accuracy.Aiming at the problem that the long-term and short-term memory networks and convolutional neural networks of the current feature extractor perform poorly in long-distance semantic capture,this thesis designs a classification model based on improved attention mechanism,and adds location information.Experiments show that improving attention The model of the mechanism is not only better than the previous model in the long-distance semantic capture problem,but also the accuracy of the classification has been improved.Finally,in order to get the emotions of the specific evaluation aspects in the user’s comment text,this thesis designs a model of the information of the hidden layer and the word embedding layer in the two-way long-term and short-term memory network model,and integrates the evaluation into the two-way long-term and short-term memory network.The method of attention mechanism can identify the emotional polarity of specific evaluation.Through the contrast experiment,the two-way long-term and short-term memory network model in the evaluation of the hidden layer and the word embedding layer achieves the best effect of the existing model.In order to identify the emotional polarity of multiple evaluations at one time,this thesis designs a multi-task learning model based on attention mechanism.Through contrast experiments,the multi-task learning model based on attention mechanism enhances the efficiency of emotion recognition and emotion recognition.The accuracy has also improved.