Uncertainty Detection in Healthcare Services: A Knowledge-fused Decision Boundary Approach against Unknown Intentions

Chatbots are widely adopted in healthcare contexts to provide real-time medical information 24/7. A formidable challenge in chatbot dialogue systems of limited language understanding ability is to handle queries with unknown intentions due to users’ diverse and uncertain consultation needs. Meanwhile, people’s asymmetric medical knowledge reserves make the process even more difficult. Failure to deal with unknown intentions may lead to a risk of incorrect or harmful information acquisition. It is necessary to empower chatbots with the capability to match medical knowledge and refuse to respond to queries with impermissible or unprocessable intentions. To address these issues, we propose a decision boundary learning approach, fusing knowledge from chatbot users and medical experts to obtain an informative query representation for unknown intention detection. The effectiveness of the proposed method is empirically validated based on real-world user query data collected from the Tianchi lab and medical data crawled from the Xunyiwenyao website.