An improvement of PhoBERT to increase the Vietnamese understanding of the hotel information chatbot

Abstract

Chatbots supporting tourism information in Vietnamese are attracting a lot of attention from researchers and businesses alike. Traditional chatbots are often built on the basis of finite rules, knowledge, or states, so they are often ineffective. Recently, thanks to the implementations of machine learning in natural language processing, chatbots have made significant strides in intent classification, entity extraction, and sentiment analysis. This paper proposes an improvement of the pre-training model to build a hotel information chatbot in Vietnamese. With the suggestion of adjusting the pre-trained models to evaluate which model works best, the simulation results show that our proposal obtained a significant effect, based on the metric for evaluating, namely, Accuracy = 96.4%; F1-score = 96.9%; and Precision = 97.4%.