Design and Development of a Sentiment Analysis System for Chinese Online Comment Texts
Tianyi Zhou, Qingchun Hu, Junzhe Li, Liuxing Wu · 2023
The limitations of manual recognition and traditional binary sentiment analysis models, such as low efficiency, low accuracy, and high cost, have become increasingly prominent and can no longer satisfy the requirements of industrial applications. In this paper, we propose a BERT-based six-class emotion analysis model that is capable of recognizing six distinct emotions, including positive, anger, sadness, fear, surprise, and neutral. Leveraging this model, we have developed an industrial-grade Chinese online sentiment analysis system that addresses the challenges of text sentiment processing in real-life scenarios. By employing knowledge distillation and model pruning techniques, we are able to enhance the accuracy and speed of the pre-trained models. This system extracts and classifies emotional information, performs retrieval and induction, and offers robust analytical indicators and sentiment classification for individuals and enterprises. The outcomes are visually presented through word clouds or graphical representations, thereby providing a solid foundation for decision-making in the relevant fields. The proposed model achieves an F1 score of 0.78, demonstrating its effectiveness and practical applicability.