.png){kind=icon}
.png){kind=icon}
Return
Towards attribute-Augmented course recommendation: An LLM-Driven model-Agnostic representation learning framework
.png){kind=icon}
PRE
AI
DOI：10.1016/j.knosys.2026.115809.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Course recommendation systems are essential to personalized online education platforms, helping learners find courses that meet their preferences and future developmental needs, thereby achieving exponential improvements in learning efficiency. Despite the promise of recent LLM advances, their integration into course recommendation remains severe challenges. First, directly encoding course titles inadequately captures their comprehensive semantic content and structured characteristics. Second, serializing lengthy textual descriptions of course histories exacerbates modeling complexity and degrades real-time performance. To address these limitations, an LLM-driven model-agnostic representation learning (LMRL) framework is proposed, which deeply integrates the powerful text generation and comprehension strengths of LLMs to enhance current course recommendation systems. For each course, meticulously crafted system prompt is employed to guide the LLM in generating fine-grained and multi-dimensional attribute information, thereby expanding the semantic space and explicitly revealing topological associations among courses. Furthermore, LMRL trains a lightweight transformation network through the constructed contrastive-aligned course representation learning module to achieve optimal modeling of course information containing multi-dimensional attributes. This approach enables LMRL to be seamlessly integrated as a preprocessing component into existing course recommendation systems, achieving substantial performance improvements with minimal computational overhead. Experiments on two education datasets confirm the efficacy and compatibility of LMRL.
Keywords:
course recommendation
large language models
representation learning
attribute augmentation
model-agnostic
Journal
K
IF:
7.6
Papers:
1.2W
Citations:
4.5W