Fine-grained Sentiment Analysis Based On User Comments

In recent years,with the continuous development of e-commerce,user reviews have become an important part of major online consumer platforms.This is because they often reflect a large amount of valuable information such as user attitudes and sentiment,and can be carried out well for business feedback.Therefore,how to efficiently extract useful information from these review texts is of great significance to consumers,businesses,and consumer platforms.Existing research mainly uses "coarse-grained sentiment analysis" and "fine-grained sentiment analysis" methods to realize sentiment tendency of text information.Coarse-grained sentiment analysis mainly calculates the overall sentiment tendency of a given text,while finegrained sentiment analysis is more specific to analyze the sentiment tendency of different evaluation objects or attributes in the text.In order to further mine the sentiment information in user comments and improve the classification effect,based on the task of fine-grained sentiment analysis,this paper proposes a text sentiment classification model combining Bi GRU-Attention and Gated Mechanisms to analyse user comments sentiment tendency.The model uses the combined features of word and character vectors as model input.Then we use Bi-GRU,to extract in-depth feature of the text.Then we combine the gated mechanisms and the attention mechanism to further extract the sentiment information related to the aspect according to the obtained aspect information;text sentiment analysis is performed on the output layer,and the final sentiment polarity is obtained through softmax.On the AI Challenger2018 fine-grained user comment sentiment analysis data set,the F1 score reached 0.7051,which exceeded the baseline system's performance.Secondly,in order to further improve the prediction performance of the model,this paper is based on the above-mentioned text sentiment classification model,uses the constructed user comment sentiment dictionary,combined with the negative vocabulary and part-of-speech features as user comment sentiment feature information,and the word vector and character vector through the Highway network to fuse.The fusion result is used as the input of the encoding layer of this model.On the AI Challenger2018 fine-grained user comment sentiment analysis data set,the F1 score reached 0.7226,the model effect has been improved,the content of feature extraction is further enriched and the performance of the model is improved.Among them,the constructed sentiment dictionary contains 5891 positive emotional words and 2963 negative emotional words.Compared with the original dictionary,the overall vocabulary has increased by 29.3%.Finally,in order to verify the effectiveness of the model in this paper,a fine-grained sentiment analysis experiment was performed on the ACSA task catering English data set,the accuracy rate reached 82.80%,and good experimental results were obtained.