MedChatBot: An UMLS based Chatbot for Medical Students

Abstract

The use of natural dialog has great significance in the design of interactive tutoring systems. The nature of student queries can be confined to a small set of templates based on the task domain. This paper describes the development of a chatbot for medical students, that is based on the open source AIML based Chatterbean. We deploy the widely available Unified Medical Language System (UMLS) as the domain knowledge source for generating responses to queries. The AIML based chatbot is customized to convert natural language queries into relevant SQL queries. The SQL queries are run against the knowledge base and results returned to the user in natural dialog. Student survey was carried out to identify various queries posed by students. The chatbot was designed to address common template queries. Knowledge inference techniques were applied to generate responses for queries for which knowledge was not explicitly encoded. Query responses were rated by three experts on a 1-5 point likert scale, who agreed among themselves with Pearson Correlation Coefficient of 0.54 and p < 0.05. The overall average rating assigned by experts was 3.4.