Automatic Riddle Generation for Learning Resources

This paper proposes a novel approach to automatically generate conceptual riddles, with an objective of deployment in online learning environments. The riddles are generated by creating triples from the learning resources using BERT language model, which are fed to the k-Nearest Neighbors language model to identify the proximity between properties and their respective contexts. These properties are classified into Topic Markers and Common based on their uniqueness and modeled on an effective instructional strategy called as Concept Attainment Model. Each riddle is passed through the Validator Module that stores all possible answers for the riddles and is used to verify the learner’s answers and provide them hints. The riddles generated by our model were evaluated by human evaluators and we obtained encouraging results.