Quick Notes - Day 13
All previous posts in this series here
theme: question generation
A Systematic Review of Automatic Question Generation for Educational Purposes
Authors: Ghader Kurdi, Jared Leo, Bijan Parsia, Uli Sattler & Salam Al-Emari
Published at: International Journal of Artificial Intelligence in Education volume 30, pages121–204(2020) url
This is a exhaustive survey of automatic question generation research specifically tuned to educational applications, and that came between 2015-2019. It gives a general overview of previous surveys of the field first and presents a detailed survey of 90+ papers that were published between 2014-19. It then discusses some challenges and future directions.
This is pretty extensive. When I started browsing, I was expecting something like a longish (15-20 pages) paper considering it only covers a 5 year period. But this is quite long, almost like a mini book, and covered almost all aspects one can think of. I liked the review objectives specification part, and description of their review method. A very useful resource to return to if one wants to work on this topic!!
Question-Worthy Sentence Selection for Question Generation
Authors: Sedigheh MahdaviEmail authorAijun AnHeidar DavoudiMarjan DelpishehEmad Gohari
Published at: Canadian AI 2020: Advances in Artificial Intelligence pp 388-400 url
This paper describes an approach to choose the sentences from which questions can be generated in a piece of text. These chosen sentences are then sent to a seq2seq model to generate questions. They show that this is a better approach than working at a paragraph level and assume all sentences as question worthy as is typical in contemporary neural approaches.
I spent some time a few years back thinking about this question, and hence, the topic appealed me. The paper takes a feature engineering approach to sentence selection, which in itself has become some sort of a novelty now. Overall, they show that this approach gives better results, but I felt there should have been one or two qualitative examples, and some discussion on what sort of questions are generated, what are not etc.