Fine-grained Sentiment Analysis Based On Part Of Speech And Location Attention Mechanism

With the rapid development of e-commerce,people are more and more inclined to buy goods and publish personal comments on the Internet.The number of comments related to goods on the Internet is increasing.Analyzing these product reviews can help businesses to find the shortage of products,and then improve their products and services,and also help consumers to better select products to meet their needs.The traditional text level sentiment analysis can analyze the overall tendency of a comment,but each aspect of a comment may show different emotional tendencies.For consumers and businesses,it is more important to master the tendentious information of all aspects of a product than the overall tendentious information.The purpose of fine-grained sentiment analysis is to obtain the information of sentiment tendency in all aspects of goods,which has important application value to help businesses and consumers better understand the information of goods.The main work of this paper is as follows:(1)Aiming at the problem that the emotion analysis model based on LSTM can't effectively mine the emotion information of specific evaluation attributes,this paper considers the influence of word part of speech and relative position information on the emotion information of mining evaluation attributes,and proposes an emotion analysis model based on part of speech and position attention: pap-lstm.The model uses part of speech information and relative position information to calculate the attention weight of words,fully excavates the influence of different words on evaluation attributes,and judges the emotional tendency of different evaluation attributes.The experimental results show that the proposed PAP-LSTM can effectively judge the emotional polarity of evaluation attributes.(2)Aiming at the problem that the traditional attention mechanism can't effectively pay attention to the context information,this paper uses convolutional neural network to capture the n-gram feature of the text,and applies it to the attention calculation of emotion analysis task,and puts forward an emotion analysis model based on CNN computing attention: cnnat-lstm.The model can capture the feature information in different context by setting different convolution kernels.By combining convolutional neural network to calculate the attention weight,the model effectively makes up for the shortcomings of the traditional attention mechanism thatcan not effectively pay attention to the context information in the calculation process,and fully excavates the influence of different words on the evaluation attributes.Experiments show that the model can generate more accurate attention weight and effectively judge the emotional polarity of evaluation attributes.