Good-Quality Question Generation for Academic Support

摘要

The paper presents a metric to automatically compute a score for machine-generated questions and transforms the questions which are having a lower score value, unacceptable, and ungrammatical into a human appealing form. Questions are unacceptable due to the flaws like incorrect grammar, selection of wrong wh-phrase, partial selection of answer phrase, negation, etc. Identifying such infirmities in the question is a challenge. Here, our attempt is to automatically detect and correct the flaws present in the question. We named this system as the Automatic Question Quality Enhancer (AQQE). By employing a multiple linear regression model, AQQE first computes the score (in range of 1-rejected to 5-accepted) for 174 questions. Higher score value tells the acceptance and lower score value shows the rejection of the question. AQQE's challenge is to enhance the quality of questions having a lower score. Out of 174, human evaluator had identified 84 questions as acceptable and 90 (51.72%) as an unacceptable. Performance of AQQE is judged with precision and recall and it is found well acceptable. AQQE enhanced 79(87.77%) questions are accepted by the human evaluator and 11 (6%) questions can be accepted with further modifications.