Comparison of Automatic Question Generation Techniques

Abstract

Automatic question generation is a technique that generates a question from various sources like structured or unstructured content. Even if, each method has its own shortcoming and strength, we didn’t find any study that did an experimental comparison of the current automatic question generation techniques using a similar dataset. In this study, we analyzed the different state-of-the-art research works, and we identified significant challenges of each model. In addition, we have used BLEU, METEOR, and ROUGE automatic evaluation metrics, to evaluate the experimental test result. From our evaluation result, we observed that most of the tested techniques score below 0.5 in all automated evaluation metrics. Out of tested techniques T5-transformer-based, scores the maximum result. This result point out this research area still needs further investigation and preparing standardized training.