Research On Fine-grained Sentiment Analysis Of E-commerce Reviews Based On Deep Learning

With the rapid development of the Internet and the popularit y of smart terminals,online shopping has become an important shopping way in people's lives.Users post comments and express emotions on various e-commerce platforms,result ing in and accumulating a large number of product review to be processed.Using natural language processing related technology to analyze these review to mine the emotional tendencies contained in them is an important way for businesses to obtain consumer feedback information.Therefore,it is of great significance to study the text sentiment analysis of e-commerce reviews.According to the size of the emotion analysis target,the emotion analysis can be divided into four levels.The size from small to large is word level,sentence level,paragraph level,chapter level.E-commerce reviews usually contain emotional tendencies towards different attributes of goods,so they are more suitable for emotional analysis at the lexical level.Lexical level sentiment analysis is fine-grained sentiment analysis.Based on e-commerce comment corpus,this paper focuses on fine-grained emotional analysis invo lving the opinion target extraction task and the related opinio n target emotional analysis task.Specifically,the research contents of this paper are as follows:(1)Study the extraction method of the opinion target.This paper presents an opinion target extraction model CNN-BiLSTM-CRF based on two-channel convo lution neural network and bidirectional long-term and short-term memory network.The model is divided into three modules: dual-channel embedding module,coding module and discriminant sequence module.The two-channel embedding module converts comment text into word vector representation and incorporates part-of-speech features.The coding module uses CNN and BiLSTM neural networks to code.The two networks capture the potential feature informat ion of text sequence in different ways,and enhance the model to extract the feature information of words around each word.The discriminant sequence module models the discriminant sequence of the feature information obtained by the coding module,and finally obtains the optimal sequence.The experimental results show that the F1 value of the model is 1% higher than that of the exist ing model.(2)Study the sentiment analysis of related opinion target.This paper propose an emotional analysis model mCNN-AttBiLSTM which combines Attent ion mechanism.Firstly,the model extracts regional features by convolut ion of convolution kernels of different sizes through mult i-channels,then fuses the word vector of opinion target with regional feature informat ion,then inputs the fused features into bidirectiona l long-term and short-term memory network to obtain context information,and adds Attention mechanism to assign different weights to the output of BiLSTM at each time.Finally,it uses Softmax classifier for emotion classification.The experimental results show that the accuracy of the model is 1% higher than that of the existing model.(3)Based on the research content of this paper,a fine-grained emotional analysis system for e-commerce is designed and implemented.