Aspect-based Sentiment Analysis And Application Research Based On Deep Learning

With the spread of the Internet and the impact of the 5G network,the proportion of netizens in our country is increasing,and a great amount of data platforms and resources have followed.For example,user preferences and opinions on e-commerce platforms and tourists’ experiences on travel websites are all invaluable research materials.Accurately obtaining valid information in the comment text can,on the one hand,help relevant departments to grasp the dynamics of public opinion and make further decisions;It can be seen that mining the key content of network text information is a very valuable task.With the in-depth exploration of deep learning and the improvement of sentiment analysis technology,research has found that aspect-based sentiment analysis can make more targeted judgments on the sentiment tendency of evaluation objects and improve the accuracy of sentiment analysis.This paper mainly focuses on the performance improvement of aspect-level sentiment analysis technology and its application.The main contents are as follows:To enhance the capability of the aspect-based sentiment analysis model,this paper proposes an interactive aspect-based sentiment analysis model(BERT-Bi-IAN),which is based on the BERT pretrained model.It dynamically represents the evaluation object and the context through the BERT pre-training model,and then collects the forward and backward semantic information of the two through the Bi-LSTM network,and inputs it into the interactive attention module to extract the interaction between the two.The reconstructed information representation serves as the input to the sentiment analysis layer.The model is applied to the Sem Eval-2014 task 4 dataset and the Chinese review datasets dataset.The experiments show that BERT-Bi-IAN can validly enhance the capability of sentiment recognition.In terms of application,this paper crawls tourist reviews about red tourist destinations on tourism websites such as Tongcheng Travel,does appropriate preprocessing of the data and extracts relevant evaluations,and finally obtains the data set required for the aspect-based sentiment analysis task,the model BERT-Bi-IAN is trained and tested on the labeled red tourism data set,and The results demonstrate that it has a an accuracy of 0.902 and an F1 value of 0.901,which is better than other classic models.Using this model and content analysis method to analyze the red tourist reviews in province A,it shows that tourists’ impression of red tourism in province A is generally positive emotional attitude,the corresponding aspects include scenic tour guides,services,scenery,etc.,and negative emotional attitudes mainly include ticket price,scenic spot management,traffic and cost performance;Neutral emotional attitudes are mainly manifested in declarative expressions and introductions to red tourist attractions.To further enhance the positive emotion of travelers and raise the image of attractions,we make recommendations to overcome the shortage of red tourist attractions in the province.