Roughly half of India’s fifth-grade students are unable to read a simple second-grade paragraph, and only a quarter can solve simple division problems. Randomized controlled trials have shown that educational programs focused on “teaching at a relevant level” (TARL) can shrink these learning gaps dramatically within a matter of months. TARL involves grouping students by learning needs rather than age in order to offer more personalized and individually relevant instruction.

Several NGOs in India are building e-Learning platforms that aim to deploy TARL at scale through interactive games, tutorials, and lessons on mobile phones. A major challenge for these platforms is to display or recommend the most relevant content to child learners who may not have the literacy skills, self-awareness, or adult support needed to select the appropriate material.

In addition, our fieldwork has shown that children in Indian classrooms are fairly likely to share user accounts in practice, which makes it necessary to re-estimate the learner’s knowledge state during each use session, and to look for multi-user patterns, rather than simply attribute an account’s full user history to one learner.

We are developing novel quantitative techniques to efficiently estimate the learner’s knowledge state using the data generated from ordinary play on the platform, even allowing for multiple learners per user account, without having to administer lengthy and formal assessments.

Our work is being conducted in collaboration with Ek Step (www.ekstep.org), an e-Learning Platform based in Bangalore. Currently, we are making arrangements to test our knowledge state inference methods vs. formal assessment tools with students in Karnataka.