A Contextualized Government Service Chatbot for Individuals with limited Information Literacy

Improving the Q&A ability of government service chatbots (GSCs) has become an important issue. In practice, a large number of users with poor information literacy often pose vague questions, which makes it challenging for GSCs to comprehend their inquiries within a specific context. In order to enhance contextualization, this study has constructed a multi-turn dialogue model that incorporates R-GCN and fuzzy logic to base on the “question-answer-context” matching process. To obtain more accurate context, we propose a re-question mechanism to further press for contextual details. Additionally, we introduce the sub-graph matching mechanism of fuzzy logic and R-GCN to improve the accuracy of implicitly representation of Chinese logic in the contextualized matching process. This mechanism allows us to prune the context-irrelevant parts in the “answer” and obtain more complete context information. We collected over 300,000 words of real cases as the test-set. The results of the experiments show that this model can significantly improve the contextualized reasoning ability of GSCs in a more humanized way. The innovative response generation method in this research, which utilizes “question-answer-context” matching, is more suitable for complex scenarios where the user may not be articulate. It helps to lower the barrier for accessing government services and provides more user-friendly assistance to individuals with limited information literacy.