Topic-oriented Opinion Recognition And Evolution Analysis In Social Networks

In recent years,with the rapid development of Internet technology and the popularity of mobile devices,users of different ages in different countries are getting used to using their own social network platforms for communication.Nowadays,social network platforms have become the epitome of the real world,and many users express their opinions on social networks.It is of great significance for enterprises to improve their products and for governments to understand public opinion by mining users' opinions in social networks.There are three main problems in current opinion recognition algorithms.First,most opinion recognition algorithms pay too much attention to deep learning technology and ignore the natural language characteristics of the text,so that the recognition results are not well combined with the semantics of the text;second,most of the opinion recognition algorithms only study from the perspective of the overall emotional tendency of the text,and do not consider the problem that different emotions will be expressed in different aspects under different topics,which makes the recognition results coarse granularity and lack of practical value;third,most of the current algorithms are limited to theoretical research,not applied to practical problems.In view of the above main problems,this paper puts forward corresponding solutions,and the main work is summarized as follows.(1)The BiLSTM-DGCN aspect-based sentiment analysis model is proposed.The BiLSTM-DGCN model proposed in this paper,based on the use of deep learning technology,combined with the dependency parsing of text,makes the model consider the natural language features of the text.Compared with other aspect-based sentiment analysis models on several public datasets,the experimental results show that the BiLSTM-DGCN model has higher accuracy and F1-score on multiple datasets.(2)The topic-oriented aspect-based sentiment analysis is realized.First,this paper collects the twitter data related to COVID-19,including 1658250 tweets and 634876 users,and annotates 5376 aspect-based emotional data for "government" and "lockdown".Then we use the existing model to perform topic detection and tracking analysis on the tweets collected in this article,and identify hot topics related to COVID-19.At the same time,we use the aspect-based sentiment analysis model proposed in this paper to analyze the "government" and "lockdown" aspect-based sentiment evolution of the data obtained in this paper,so as to realize topic-oriented aspect-based sentiment analysis.(3)A visual interactive system is designed and developed to integrate all the functions of this paper.The system integrates the data acquisition function,supports the acquisition of specific keyword tweets and user information,and persisting the crawling results in the Mongo DB database.The system can perform preprocessing operation on the specified data,and carry out topic detection and tracking,opinion identification on the data in the database.The system can visually display the results of topic detection and tracking and opinion recognition.