BiGRU-Attention Sentiment Analysis for Enhancing Task-Oriented Chatbots
Nguyen Thi Mai Trang, Nguyen Ngoc Hung · 2024
Nowadays, chatbots have become increasingly popular for supporting users and addressing inquiries. However, chatbots still cannot entirely replace humans in many cases, such as handling complex questions or understanding users’ emotions through messages and emoticons. Therefore, understanding user expressions through their statements is an essential task to enhance the quality of chatbots. This study introduces a method to enhance a task-oriented chatbot designed for the Vietnamese language by incorporating a deep learning-based sentiment analysis model into the system. In this research, the proposed sentiment analysis model emphasizes the inclusion of bidirectional gated recurrent units (BiGRU) and an attention layer in the model architecture. Subsequently, the task-oriented chatbot utilizes the created model to comprehend user expressions in situations of feedback intent. The model has achieved an AUC of 93.91% on the UIT-VSFC test set for classifying three labels: positive, negative, and neutral. Applying sentiment analysis models to chatbots helps them interact with users more naturally and automatically gather detailed feedback on user satisfaction, frustration, and specific areas for improvement.