Aspect-level Sentiment Analysis Of Internet Reviews Based On Deep Learning

Internet review is an important carrier for people's opinions today.People often express opinions on many aspects when commenting.Aspect-level sentiment analysis of internet reviews is a fine-grained task that extracts the opinions and emotions of the opinions contained in internet reviews.The goal of this paper is to mining the various aspect terms and related opinion terms contained in the reviews and classify the sentiment polarity contained in each aspect term.This paper divides the sentiment analysis into the following two steps:(1)Aspect-level sentiment element extraction based on Tree-LSTM+CRF.In this part,we need to extract Aspect-level sentiment elements from internet reviews and get the aspect terms and opinion terms contained in the reviews.In this paper,a Tree-LSTM+CRF joint model is proposed to extract aspect terms and opinion terms.The model first constructs hidden vectors of tree-structured long short-term memory neural network learning words based on dependency parsing tree,and then uses conditional random fields for sequential tagging.(2)Classification of sentiment polarity based on Attention Bi-LSTM.In this part,we need to classify the sentiment polarity of the extracted aspect terms.In this paper,Attention Bi-LSTM model is proposed to classify the sentiment of the aspect terms.The model learns the representation vectors of each word through two-way long short-term memory neural network and uses attention mechanism to learn the correlation between words.The model can not only realize Aspect-level sentiment classification but also can classify the words according to the attention weight.The aspect term is matched with the content of the relevant opinion term.This paper implements many baseline models for sentiment element extracted and baseline models for sentiment classification and compares these baseline models to the two models presented in this paper on the data set of task 4 of SemEval Challenge 2014.The experimental results of the comparative analysis show that the performance of this model is better than that of the baseline model.The validity of Tree-LSTM+CRF model and Attention Bi-LSTM model in sentiment element extraction and sentiment classification is verified.