An Extendable Sentiment Monitoring Model for SNS Considering Environmental Factors

Rapid growth of social network service (SNS) has drawn significant attention from the publics. Existing research indicated that emotional state and behavioral tendency of SNS users can be identified and predicted through sentiment analysis. However, it is found that not only the posts can express user’s emotions but the overall environmental conditions faced by the user may lead to the generation of different emotions based on the cognitive theory of emotion and observation. Therefore, it may lead to bias between the sentiment analysis results and the actual situation if only analyzing the post. This study targets to propose an extendable sentiment monitoring model which considers the actual environment of users in SNS. Through this model, the result of sentiment analysis is closer to reality. By analyzing the content of users’ continuous posts, the sentiment analysis can take into account the pre- and post-textual relationships. The classification result of external affecting sentiment factors by K-means is used as criteria for weighting method to adjust the results of sentiment analysis based on BERT. Finally, the time series analysis is used to predict sentiment tendency monitor sentiment changes. The experiment results show that the training and validation accuracy are 89.24% and 84.00%, respectively. By our weighting method to revise the BERT results, the F1 score is improved from 0.839 to 0.850.