BICASH: BERT-based Integrated Analysis of Campus Sentiment with Sequential Histories

Abstract

In the era of digital communication, social media and online platforms have become prevalent channels for expressing emotions and opinions, particularly in campus communities where students frequently share their daily lives, learning experiences, and emotional states. Accurately identifying the sentiment of students’ comments is crucial for analyzing their current psychological state. However, traditional sentiment analysis methods mainly focus on the explicit content of texts, often overlooking the potential impact of a user’s past emotional expressions on their current emotional state. Therefore, this paper introduces BICASH, a novel sentiment analysis method based on user historical sentiment analysis. This approach assists in more accurately judging the sentiment of current comments by using the BERT model for preliminary analysis of current sentiment, and the LSTM model to capture the sequential relationship between historical comments, thus refining the assessment of current sentiment. We conducted experiments with this model on a campus comment sentiment analysis dataset. The results show that the BICASH model, utilizing 50 historical comments for sentiment extraction, achieves the best performance with a precision rate of 0.8311, recall rate of 0.7943, and F1 score of 0.8123, outperforming other baseline models.