EdNet

EdNet is a dataset of learner interactions collected by Santa, a multi-platform AI tutoring service for the TOEIC English proficiency test, built by the Korean education technology company Riiid. It gathers 131,441,538 interactions from 784,309 students over roughly two years, spanning 13,169 skill-tagged questions and 1,021 lecture videos, with each question mapped to one or more of 293 skill tags. Every logged action carries a timestamp, so the data captures not just whether an answer was correct but the sequence and timing of how a learner worked through problems, read explanations and watched lectures.

Access is open and free. The GitHub repository documents the schema and links to downloadable archives of four hierarchical subsets, KT1 to KT4, that add progressively more detail: KT1 holds question-and-response logs, while KT4 adds problem-solving action sequences, lecture and explanation activity, and purchase events. Data ships as per-student CSV files, and the subsets range from a few gigabytes each, so the whole collection is straightforward to process on a single machine.

For an AI tutor this is a behavioural corpus rather than a content corpus. It will not teach a model any TOEIC material, but it is the richest public record of how real learners progress, which makes it the reference dataset for knowledge tracing: predicting a student's future performance from their past interactions, estimating mastery of individual skills, and sequencing practice to target weak areas. That complements the teaching content in sibling sources such as openstax, khan-academy, ck-12 and siyavula, and the structured curriculum standards in common-core and ngss: those supply what to teach, while EdNet informs when and in what order to teach it.

The licence is Creative Commons Attribution-NonCommercial 4.0, released for research purposes. Attribution is required and commercial use is prohibited, so it is unsuitable as the basis of a paid tutoring product without separate permission from Riiid. As learner-interaction data it also warrants care around privacy, even though identifiers are anonymised.