Research And Implementation Of Sentiment Analysis Based On User Comments

The development of Internet technology has led to an exponential increase in user comment data.How to extract effective information from massive data is an important hot spot in current Internet technology research.And sentiment analysis technology is used to deal with and solve such problems.This paper mainly studies the fine-grained sentiment analysis based on user reviews,and analyzes the sentiment tendency of users to restaurant reviews from multiple aspects and different levels.Fine-grained sentiment analysis has also made certain research progress in recent years,and it also reflects its application value and broad application prospects in practice.Aiming at the sentiment analysis problem of user comments,this paper proposes an sentiment analysis model based on the attention mechanism and an sentiment analysis model based on BERT.The main work of this paper is as follows:1.Use XGBoost algorithm for feature extraction.Using XGBoost has the characteristic of sorting features,each fine-grained classification task is performed,and the feature importance ranking result is obtained,and finally each fine-grained aspect feature words are obtained.2.Propose the BiLSTM-Att-GloVe model,which uses the bi-directional long short-term memory network to encode the context and feature words respectively to obtain the feature vector representation,and then obtain the attention weight of different words in the context for the aspect words through the attention mechanism,finally,Softmax is used to obtain the probability distribution of the emotional tendency of the text.Because BiLSTM cannot achieve parallelism and has a gradient problem,the Att-Con-GloVe model is proposed based on the attention mechanism,which uses the multi-head attention mechanism to encode feature words and text,thereby extracting the vertical space feature vector of the text.Experiments prove that the Att-Con-GloVe model is better.3.The BERT model proposed in recent years ranks first in many fields,so it is considered to be used in the task of this article,mainly in two ways.The first is to use the BERT-based sentiment analysis model,use the sentence-level output of the BERT model,and connect to the sentiment analysis classification task,and fine-tune the pre-training model to realize the fine-grained sentiment analysis task.The second is to propose the Att-Con-BERT model,which uses the word-level output of BERT to embed the BERT model as a word vector of the text,and then enters the model based on the multi-head attention mechanism and convolution transformation.Experiments have proved that the accuracy of the BERT-based model is higher,and the Att-Con-BERT model has the best effect.4.Design and implement a user comment system based on sentiment analysis,which can directly call the trained sentiment analysis model to obtain the sentiment analysis results of user comments.