• Technical Name
  • Automatic Reading Comprehension Question Generation based on Deep Learning Language Model
  • Operator
  • National Chung Hsing University
  • Booth
  • Online display only
  • Contact
  • 范耀中
  • Email
  • yfan@nchu.edu.tw
Technical Description This technology is a natural language generation (Natural Language Generation) model based on deep learning, which can generate grammatically fluent sentences for reading comprehension test preparation. The completion of this technology can significantly reduce the editing labor cost of the publishing industry or marketing business in reading inspection. The quality of questions generated by the model has reached the level of human quality.
Scientific Breakthrough At present, the quality of questions generated by our model has reached the quality of human-generated questions. For experimental evaluation, we use the SQUAD2.0 data set released by Stanford University. We use BLEU 4 Score as a performance indicator. Our first question generation model score reaches 22.3, which was the SOTA result in 2019. At present, in our latest research results, we further advance the BLEU4 Score to 39.5, which again is a SOTA result in 2020 (papers submitted to EMNLP2020 for review). Based on this technology, we further investigate different question type generation: such as multiple distractor generation and MCQ summary-like answer generation techniques.
Industrial Applicability  Target market: Education industry and Publishing industry. To shape the development of education technology (EdTech) industry
 Existing mechanism: Existing industries mainly rely on manual methods for generating questions and test questions or question bank methods, which have the disadvantages of slow speed and high cost
 Reshaping mechanism: automatic generation technology can assist teachers or editors to improve efficiency and teaching quality.