Building Filters for Vietnamese Chatbot Responses

Automatic Chatbot is a developing trend in the world because of its convenience, efficiency, applicability in many fields such as E-commerce, customer care, and so on. Through Chatbot, we can cut down human resources at the place that Chatbot can replace, increase the convenience for both customer and company. Some good chatbot systems in the world primarily used in English. With Vietnamese, however, the outputs of these models are often not so good due to different structures and grammar. Some output (i.e. chatbot's responses) is incorrect or even misleading to the user. This can cause some serious problems in some legal or financial applications. In this paper, we proposed an approach to reevaluate outputs of general chatbot to reduce incorrect responses. Specifically, we firstly deployed well-known approaches in English chatbot into Vietnamese chatbot. After that, we build filters to assess chatbot responses based on the characteristics of the Vietnamese language. If the response is not suitable, the system will not give back to the user. In this case, the system can require more information to find a better response or at least return announcement instead of an incorrect response. We also developed a web-based application with chatbot integration to evaluate our approach in real life scenario.