Commodity Feature Extraction And Sentiment Analysis Based On Consumer Comments

When purchasing products on e-commerce websites,users first check the description information of merchants,and also check the comments of historical consumers.This kind of comment information can guide user's consumption behavior,and is the key factor affecting user's final consumption.However,with the rapid growth of the number of comments,it is difficult for users to easily and quickly grasp the commodity features and sentiment from historical consumers' s comments.Therefore,it is particularly important to intelligently analyze comments and perform fine-grained sentiment analysis on commodity features to help users understand a large number of consumers' s comments.The main research work of this thesis is based on the deep learning method,which automatically extracts the commodity features from consumer comments,and conducts sentiment analysis on this basis.The specific contents include:1.Construct the CAtt-BiLSTM-CRF neural network model to extract the opinion target and opinion expression of commodity comment corpus.In this model,a bidirectional long short-term memory network based on attention mechanism is used to automatically obtain the context information of the words in sentences.It does not need to pre-analyze the syntactic dependencies and manually construct feature engineering.Finally,the past and future labels are used to predict the current time label through linear chain conditional random field,thereby improving the accuracy of opinion target and opinion expression extraction.2.Construct models to extract opinion relation,cluster typical opinion target and predict the opinion expression's polarity,and then achieve fine-grained sentiment analysis based on opinion target.Firstly,opinion relation is extracted by relational classification method,so that independent opinion target and opinion expression can be paired.Then,the opinion target are clustered into several typical features by clustering algorithm Vec-HC,and opinion expression's polarity is classified by classifier.3.Design and implement an opinion target extraction and sentiment analysis system based on consumer comments,apply the models proposed in this thesis to analyze consumer comment data,extract commodity features and display fine-grained sentiment analysis.