Students learn more efficiently when they are confidently engaged. And deep learning affords such engagement.
Global AI tutor provider and a member company of the Born2Global Centre, Riiid shows why in a recent real-world study demonstrating how personalized intelligent tutoring systems (ITS) predicting precise test results with deep learning algorithms can boost student enthusiasm, energize motivation, and will eventually fuel further business growth.
The paper "Prescribing Deep Attentive Score Prediction Attracts Improved Student Engagement" was recently accepted by top-tier education gathering EDM 2020 (Educational Data Mining), the 13th annual international conference on data mining, happening virtually July 10-13, 2020.
Conducting a detailed A/B test featuring two models – a collaborative filtering algorithm and a deep-learning algorithm – Riiid researchers examined about 78,000 users and their accompanying indexes indicating motivation and engagement. Those factors included test completion rate, registration rate, number of questions solved, conversation rate, and average revenue per user.
Every index showed consistent improvement with the deep-learning algorithm.
Riiid's deep-learning score prediction model tested in this research runs with "assessment modeling," the fundamental pre-training tasks for interactive educational systems. The firm's researchers proposed the pre-train and fine-tune methods of predicting exam scores with student-system interaction data when label data - the actual exam score data in this case - is scarce.
This new scores prediction model showed 36.8% higher accuracy compared to the previous model in terms of MAE (Mean Absolute Error).
The study concludes perceived precision of predicted scores increased test-takers' confidence in solutions, which in turn encourages more active engagement.
One reviewer said, "This is an excellent, practical example of closing the loop from model prediction to actual tangible benefits."
The study was one of the few done prospectively in a real-world setting, as opposed to retrospectively. The results are more realistic when taken outside a laboratory setting, providing more of a tangible value than mere theory.
"Even for AI scientists, most of the time, it is challenging to answer what areas a certain deep learning technology can be implemented in and drive meaningful impact," said Youngnam Lee, Riiid's researcher and the lead author. "This research was only possible because functions of AI research and product operations are tightly aligned and work together at Riiid."
Riiid deployed a deep-learning prediction model to all the users, expecting a higher level of user engagement bringing improved learning efficacy.
Putting utmost efforts in R&D to advance its AI models to enhance their functions, Riiid strives to earn users' trust and drive active engagement - both of which are direct indicators of business growth, as this study shows.
"Riiid's AI doesn't just present a better way of learning; it has great measurable value for business," said YJ Jang, CEO and co-founder of Riiid. "We will continue to study and research fundamental AI tech that brings real-world industrial impact proven through decisive and rapid deployment. We will keep creating unprecedented cases appreciated by the business and academic communities."