PhD Research Seminar: Ontology-Controlled Automatic Item Generation in Personalized Learning Environments, Machine Reading Comprehension with Interpretable Discrete Reasoning
First talk: Methods and Software Tools for Ontology-Controlled Automatic Item Generation in Personalized Learning Environments
Speaker: Mariia Gordenko, second-year PhD student, Faculty of Computer Science
In the last decade, there has been a great increase in the number and variety of methods for automatic item generation (AIG). It can be partly explained by the infiltration of educational data mining methods into the traditional AIG approaches mainly based on item response theory, testlet response theory, and other well-known testology models. We overview ontology-based methods for generating and evaluating the quality of assessment items and propose new approaches for generating open-ended questions. We discuss the differences between the traditional AIG and intelligent methods based on data science, as well as examples of using ontology-controlled AIG methods.
Two use cases are considered. Firstly, COVID-19 highlights not only the strengths but also the weaknesses of existing educational software. The lack of an engineering approach in the deployment of distance-learning tools leads to technological and methodological problems. Significant ideological differences between educational software generations are discussed with special attention to non-cloud-based and cloud-based collaborative technologies and corresponding platforms. Thus, the power of integration based on industry-wide interoperability standards can help solve current problems in distance education related to software.
Secondly, we observe an increasing number of users of corporate educational solutions utilizing cloud architecture. However, non-cloud-based learning tools do not meet the requirements imposed by this growth. We name technologies and product features that allow corporate solutions to quickly gain popularity among educational society. We provide clear examples of their connection to learning methods that can improve teaching, learning, and last, but not least, user’s experience. Finally, we highlight the significant role of integration and interoperability standards supporting easy components replacement and scaling.
Second talk: Machine Reading Comprehension with Interpretable Discrete Reasoning
Speaker: Ramil Yarullin, second-year PhD student, Faculty of Computer Science
Machine reading comprehension aims to enable machines to answer questions over a given context. Inferring the answer may require multiple-step reasoning and involve discrete operations such as extracting spans from a text, negation, counting, numerical comparison, sorting, and arithmetic expressions. Recent works also show that even without additional supervision, it is possible to train neural networks to output the reasoning process in the form of a program whose execution against a given context yields a correct answer. In this talk, we will overview the existing approaches in this line of research and discuss possible ways of creating a model with both interpretability and state-of-the-art performance.