Thai Question-Answering System Using Pattern-Matching Approach

Teaching concepts in Thailand's universities have abruptly changed, due to the advancement of the COVID-19 pandemic, including changes in classroom to online formats, as well as administrative difficulties. The research herein, therefore, addresses these concerns, presenting a Thai question-answering system using the pattern-matching approach. Our case study covers course information, teaching timetable, teacher schedule, and course supplements. We classified the questions into six categories according to type and acknowledged typical expressions which matched to question patterns. We use RegEx® to match a defined pattern. When a response did not match, we used word embedding to transform the question into a vector and then calculated the cosine similarity to identify the most similar pattern. The system can then generate a corresponding SQL command to query the answer from the database. We evaluated the accuracy of the proposed system with the collected data resulted in an accuracy rate of 82%.