Design And Implementation Of Fine-grained Sentiment Analysis System For Commodity Reviews

Traditional text sentiment analysis has shown a good value in some fields that only need to determine sentiment polarity,such as online public opinion analysis and stock evaluation analysis.However,with the further development of application,traditional sentiment analysis cannot fully meet the demand if users want to obtain the specific sentiment analysis results corresponding to the attributes of the evaluation object.For this reason,fine-grained sentiment analysis came into being and got more and more attention.However,there are still many difficulties and challenges in the text representation of finegrained sentiment analysis.For example,product reviews are web-based texts,which have features such as short text,too much key information,irregularities,and complicated expressions.There may even be problems with unregistered words or emotion words that lack object attributes.In addition,the existing e-commerce platform has not yet realized the practical application of the fine-grained sentiment analysis system.How to apply the algorithm to real life is also a challenging task.This article focuses on these key issues in fine-grained sentiment analysis,based on the summary of domestic and foreign fine sentiment analysis theories and research results,the main research content includes the following five parts:(1)By introducing Term Frequency-Inverse Document Frequency(TF-IDF)technology to weight the keyword vectors,and then use Text Convolutional Neural Networks(TextCNN)for emotion classification,Thereby improving the accuracy of fine-grained emotion classification.Experiments show that the accuracy of fine-grained emotion classification is improved compared to traditional TextCNN,which provides a good solution for solving the problem of keyword vector weighting.(2)Strengthen the text representation by introducing joint word representation during input,and add an extra layer of long-term and short-term memory models before classification using the gate-oriented convolutional network with aspect embedded(GCAE)model(Bi-Long Short Term Memory,BiLSTM)improves the problem of long sentence information loss,thereby improving the accuracy of fine-grained emotion classification.Experiments show that the accuracy of fine-grained emotion classification is improved compared to the classic GCAE model,which provides a good solution for improving the classic model.(3)To address the shortcomings of static word vectors in text representation,the dynamic word segmentation pre-training model is used to enhance text representation,and the pretraining model(Bidirectional Encoder Representations from Transformers,BERT)is used for fine-tuning to achieve a better effect of text expression.And use the integrated learning classification method to strengthen the sentiment classification of the two classifiers,so as to achieve a more accurate and fine-grained sentiment analysis.Experiments show that the model has the best effect and precision for improving the precision of fine-grained sentiment analysis.(4)In response to the irregularity of product review texts,this article uses an improved model based on deep learning to enable it to automatically learn features of new vocabulary without additional human intervention.(5)This thesis encapsulates the model with the best precision accuracy of fine-grained sentiment analysis into a prototype system,describes the requirements analysis of the prototype system,and implements the visualization of the results of fine-grained analysis,laying a foundation for the practical application of fine-grained sentiment analysis system.In short,this thesis proposes some technical measures for the inherent shortcomings of the product review web text.From the perspective of improving text representation and optimizing the classifier,the accuracy of sentiment classification is improved.In addition,the model with the best accuracy of sentiment analysis proposed in this paper is packaged into a prototype system,which realizes related functions from the perspective of system design,which can provide a reference for the practical application of finegrained sentiment analysis system.