Design And Implementation Of Sentiment Analysis System For Product Reviews

With the development of e-commerce,the number of product reviews is increasing exponentially on many online shopping platforms.These massive review data contain a lot of users' emotional information,which has high application value.The traditional sentiment analysis technique directly gives the overall emotion tendency of the evaluation statement,while most users always pay more attention to the sentiment polarity on a specific aspect of the product.When there are multiple aspect terms in the evaluation statement,and the corresponding sentiment polarity conflicts,the traditional analysis method cannot make a correct judgement.The finer-grained sentiment analysis technique can dig out multiple aspect terms contained in the comment statement and predict the corresponding sentiment polarity for each aspect term.Therefore,it is of great significance to study the fine-grained sentiment analysis technique.In this thesis,two sub-tasks of fine-grained sentiment analysis,namely aspect term extraction and sentiment polarity prediction,are studied in depth.On this basis,a fine-grained sentiment analysis system for product reviews is implemented,and the specific work is as follow:(1)The Bi GRU-CRF model was constructed for the aspect term extraction of evaluation reviews.The model takes Word2 vec with simple features and domain-related features as input.The hidden layer uses bidirectional GRU model to fully capture the dependencies between the context of the evaluation statement.Finally,the conditional random field model is used as the output layer to mark the output sequence.The dependence between past and future tags is taken into full account in the prediction of current tags,thus improving the accuracy of aspect term extraction.(2)The TC-Bi GRU-Attention model was constructed for the task of predicting the sentiment polarity on aspect terms.This model makes appropriate adjustments on TC-LSTM model,which splicing the mean word vector corresponding to the aspect term and the word vector corresponding to the evaluation statement as input.The hidden layer adopts bidirectional GRU to extract the semantic feature information of the context of the evaluation statement and introduces the attention mechanism to obtain the long distance dependency of the evaluation statement.According to the influence of the context on the prediction of the sentiment polarity on the aspect term,the corresponding weight is given,which solves the problem of insufficient feature information capturing and improves the prediction effect of the model.(3)A fine-grained sentiment analysis system for product reviews is designed and implemented on the basis of aspect term extraction and sentiment polarity prediction tasks research.The system visually displays the proportion of the aspect terms and sentiment polarity in the form of pie chart.Finally,the function and performance of the system are tested,and the test results show that the system meets the basic needs of users.