Which factors matter in data-driven peer learning: Social intimacy, proficiency gaps, or selection autonomy?

Peer learning is a promising strategy that encourages knowledge co-construction and reflection, while biased partner choice may hinder effective collaboration. Data-driven systems with interpretable learning analytics create new affordances for optimizing peer selection through knowledge awareness. Grounded in a technology-enhanced learning environment, this study investigates how social intimacy, knowledge proficiency gaps, and partner-selection autonomy affect peer learning in a Japanese university academic reading course. The data-driven conditions that underscored proficiency gaps in partner selection mitigated the influence of social intimacy and improved efficiency without compromising learning outcomes, although they alone were insufficient to inherently enhance engagement or learning quality. Helper-initiated pairing demonstrated relatively higher efficiency when learners could autonomously identify knowledge gaps and select partners. This study contributes a substantial implementation of computer-supported peer learning, an empirical analysis of key determinants of peer interaction, and a pedagogically aligned framework for integrating data-driven support into higher education to promote collaborative learning experiences. It also leads to future exploration of balancing these factors through contextualized design and continuously updated learner models incorporating multimodal data, thus extending applications to broader educational contexts.