PolyAQG: An Automated Question Generation Framework Based on a Combination of Four Analytical Approaches

摘要

Automated Question Generation (AQG) that automatically generates questions from text effectively reduces the time and effort of educators in preparing and setting assessments questions. Since the 1970s, AQG researchers have proposed and applied a vast range of approaches to generate questions. With that, the PolyAQG Framework is proposed in this paper by combining four analytical AQG approaches into a framework to increase the number, the variety, and the quality of questions. The novelties offered by this PolyAQG Framework are highlighted in this paper. A prototype is developed under JavaFX platform and integrated with the Stanford NLP parser. From the 300 test sentences, the framework has successfully generated ten-times more questions with more than 85 of the generated questions were rated accurate and relevant. This test result revealed that the framework did successfully meet the three research objectives formulated for this study, as follows: First, a framework is developed by combining four analytical AQG techniques. Second, two phases are used to generate new sentences from the input and earlier phases are re-used to develop a new set of questions. Lastly, the generation of ontology-based questions phase does not merely generate distractors (typical in other ontology-based AQGs), but instead generates new sentences with ontology knowledge and uses them to create a new set of questions.