Fine Grained Sentiment Analysis On Product Reviews

With the continuous development of e-commerce platforms,a large number of netizens' comments on commodities have been accumulated on the Internet.These comments not only provide emotional feedback for e-commerce users,but also provide reference for many netizens' decision-making in shopping.In the field of natural language processing,Chinese text emotion analysis algorithm can analyze users' positive,negative or neutral emotional attitudes from a large number of comment data.However,we not only want to understand the user's overall emotional feedback on the product,but also need to further understand the user's attitude towards specific aspects of the product.In order to solve the above problems,this paper further studies the model methods involved in fine-grained emotion analysis for e-commerce user product reviews based on previous studies.The core task is to extract the subject words and emotion words in user reviews in pairs,and make a judgment on the emotional polarity of reviews.The specific work is as follows:Firstly,a recognition model of subject words and affective words is studied and implemented.Length in the proposed hybrid model,two-way memory networks can make full use of context information and grasping the global features,part of speech of characteristics of attention mechanism network that will be the part of speech feature vector length and two-way memory network hidden layer and output vector calculation of attention for the part of speech contribution to predict target matrix and conditions with the airport layer can effectively learning before and after the dependencies between the output label,then determine the keywords and the emotion words in the evaluation of the user.Experiments show that the hybrid model can effectively improve the recognition effect of subject words and affective words.Secondly,a pair relation extraction model of convolutional neural network based on multivariate feature fusion is studied and implemented.This model can effectively integrate syntactic relation features into deep learning network and solve its distributed semantic representation.In addition,the high-dimensional abstract features extracted by the convolutional neural network and the artificially defined features are effectively spliced to form multiple features,so as to jointly participate in the calculation of relational extraction.Experiments show that the convolutional neural network model with multi-feature fusion has good extraction performance in relation extraction.Thirdly,a fine-grained affective polarity judgment model is studied and implemented.This model aims at the inaccuracy of emotional polarity judgment caused by semantic transition in the user's product reviews,and designs a fine-grainedemotional polarity judgment framework for different topics of the product evaluation itself.The experiment shows that the model improves the accuracy of the judgment of emotional polarity,eliminates the ambiguity to the greatest extent,improves the influence of unknown words on the judgment of emotional polarity,and achieves a high accuracy rate.On the basis of the above work,this paper also designs and implements a product review-oriented fine-grained sentiment analysis system prototype for result verification and application.