RIGL: A Unified Reciprocal Approach for Tracing the Independent and Group Learning Processes

TL;DR

The paper tackles holistic knowledge tracing by jointly modeling independent and group learning and their reciprocal influence. It introduces RIGL, a three-part architecture with a time frame-aware reciprocal embedding module, a relation-guided temporal attentive network, and a bias-aware contrastive objective to stabilize training. Across four real-world datasets, RIGL significantly outperforms baselines on both individual- and group-level predictions, with larger gains at the group level, and provides insights into how individual learning shapes group performance and how inter- and intra-group relations evolve over time. The work advances practical tools for dynamic, dual-level educational assessment and offers a foundation for uncovering latent structures within collaborative learning settings.

Abstract

In the realm of education, both independent learning and group learning are esteemed as the most classic paradigms. The former allows learners to self-direct their studies, while the latter is typically characterized by teacher-directed scenarios. Recent studies in the field of intelligent education have leveraged deep temporal models to trace the learning process, capturing the dynamics of students' knowledge states, and have achieved remarkable performance. However, existing approaches have primarily focused on modeling the independent learning process, with the group learning paradigm receiving less attention. Moreover, the reciprocal effect between the two learning processes, especially their combined potential to foster holistic student development, remains inadequately explored. To this end, in this paper, we propose RIGL, a unified Reciprocal model to trace knowledge states at both the individual and group levels, drawing from the Independent and Group Learning processes. Specifically, we first introduce a time frame-aware reciprocal embedding module to concurrently model both student and group response interactions across various time frames. Subsequently, we employ reciprocal enhanced learning modeling to fully exploit the comprehensive and complementary information between the two behaviors. Furthermore, we design a relation-guided temporal attentive network, comprised of dynamic graph modeling coupled with a temporal self-attention mechanism. It is used to delve into the dynamic influence of individual and group interactions throughout the learning processes. Conclusively, we introduce a bias-aware contrastive learning module to bolster the stability of the model's training. Extensive experiments on four real-world educational datasets clearly demonstrate the effectiveness of the proposed RIGL model.

Figures (8)

• Figure 1: An illustrative example of the holistic knowledge tracing (HKT) task. The top and bottom halves indicate the individual and group learning processes, respectively, which are organized in time frames, and the radar chart in the middle represents the knowledge proficiency levels of both.
• Figure 2: The overview architecture of our proposed RIGL model. (a) The time frame-aware reciprocal embedding module includes both individual and group interaction modeling as well as reciprocal enhanced learning. (b) The relation-guided temporal attentive network models the complex learning processes with dynamic changing knowledge, including the relation-guided dynamic graph modeling and a temporal self-attentive network. (c) The contrastive learning module generates the augmented student interactions by randomly flipping responses considering the learning bias during exercise solving, such as carelessness or guessing, and promotes the training stability through this bias-aware contrastive learning. Best viewed in color.
• Figure 3: Performance of ablation studies conducted on four datasets, where “w/o” means removing the target module.
• Figure 4: Results of reciprocal effect study conducted on NIPS-Edu dataset, where “w/o” means removing the target feature.
• Figure 5: Sensitivity analysis of learning rate on SLP-Bio.