HA-LPR: A highly adaptive learning path recommendation

Guoqian Luo, Hengnian Gu, Xiaoxiao Dong, Dongdai Zhou

Abstract: In the realm of e-learning, supporting personalized learning effectively necessitates recommending sequences of learning items that maximize learning efficiency while minimizing cognitive load, all tailored to the learner's goals. These recommendations must account for the prerequisite relationships among learning items and the learner's characteristics to create highly adaptive learning paths. Current methods encounter two primary challenges in recommending such adaptive learning paths. Firstly, they fail to formulate reasonable learning goals by integrating established learning principles. Secondly, they struggle to balance the trade-offs between improving learning performance (P), reducing learning time (T), and smoothing the transition of learning item difficulty (D), hindering optimization for both high learning efficiency and low cognitive load. To address these issues, we propose a novel Highly Adaptive Learning Path Recommendation model (HA-LPR). Specifically, the learning goals and characteristics acquisition component first determines each learner's knowledge level based on their historical interaction records, then formulates reasonable learning goals by considering their knowledge level, historical interactions, and the prerequisite relationships among learning items. The learning items recommendation component generates a candidate action space aligned with each learner's goals. It then selects an action from this candidate space that maximizes the overall P-T-D gain in the learner’s learning path by employing a multi-critic actor-critic reinforcement learning network. Extensive experiments on three real-world datasets demonstrate that HA-LPR significantly outperforms current state-of-the-art baselines.

External IDs:dblp:journals/eait/LuoGDZ25