Riiid’s AIEd Research on Strengthened Accuracy of Student Assessment and Short Answer Applications Accepted By AAAI 2022

Riiid, a leading AI Education solutions company, announced that its new paper demonstrating a multi-task learning (MTL) framework that combines knowledge tracing (KT) and option tracing (OT) models for more precise student assessment was accepted by the Association for the Advancement of Artificial Intelligence (AAAI) for its annual conference in February. Riiid will present the key findings of the paper virtually during the 36th AAAI Conference held between Feb. 22 and Mar. 1.

Developed independently by a group of in-house researchers, Riiid’s new learning framework tackles the limits of both the conventional deep learning-based KT approach that only considers whether the student has answered a question correctly or incorrectly, and the constraints of OT models that study student’s option choice to multiple-choice questions but fails to take correctness into account.

Recommended AI News:  CharliAI Launches First-Of-Its-Kind Intelligent Content Management Platform for Enterprises

“What we propose in this paper is a breakthrough Dichotomous-Polytomous Multi-Task Learning (DP-MTL) model that simultaneously learns to predict both the student’s correctness and option choice for a given question. It’s a novel architecture design that integrates KT and OT into existing KT models,” said Jason Juneyoung Park, AI Research Lead at Riiid. “To the best of our knowledge, this is the first work that combines KT and OT methods for better student assessment purposes.”

Riiid’s AAAI paper “No Task Left Behind: Multi-Task Learning of Knowledge Tracing and Option Tracing for Better Student Assessment” confirmed that the consolidated method improved KT, OT and score prediction (SC) performances about 10% versus the popular models that use the KT methods only.

Riiid’s DP-MTL model uses the option embeddings directly to output logits for each option, independent of positions. Such architecture not only improves the accuracy of Riiid’s previous score prediction model centering around the multiple-choice format but also strengthened its existing solution targeting those in the short answer format.

“The new KT-OT consolidated architecture allows our AI model to study the student response data in short answer format and predict performance more precisely. We are also greatly honored that the three peer reviewers commented that our research was among top 25% of all papers accepted to AAAI and that it will ‘surely be popular’ in years to come,” said Suyeong An, Riiid’s AI researcher and co-author of the paper.

Recommended AI News: Creator Economy Platform Koji Announces $20 Million Series B Fundraise

Riiid has been active in publishing academic works at top global conferences such as the Conference on Neural Information Processing Systems (NeurIPS) since 2016. Last year alone, the global AI and education technology conferences including the International Learning Analytics and Knowledge Conference (LAK), the International Conference on Artificial Intelligence in Education (AIED), and International Conference on Educational Data Mining (EDM) accepted six papers submitted by Riiid researchers.

“Research sits at the very core of Riiid’s mission to serve every learner’s distinct needs with personalized AI solutions. We will continue to lead the agenda of AI-enabled learning experience in the education and training sectors with our world-class, pioneering academic works. The new DP-MTL framework will be a key stepping stone for Riiid to provide more solutions in more academic disciplines and in different stages of learning,” said Riiid CEO YJ Jang.

Recommended AI News: HARMAN Acquires Mixed Reality Pioneer Apostera

[To share your insights with us, please write to sghosh@martechseries.com]