AGNE: Attentional Graph Convolutional Network Embedding for Knowledge Concept Recommendation in MOOCs

Jiahui Chen, Dan Meng, Xiangyun Gao, Liping Zhang, Chao Kong

Abstract: The knowledge concept recommendation aims to identify and suggest specific knowledge concepts that a student needs to master on Massive Open Online Courses (MOOCs) platforms, addressing the information overload issue and offering a tailored educational experience. Existing studies have constructed heterogeneous information networks to capture accurate representations of both learners and concepts, thereby mitigating the challenges posed by data sparsity. However, these approaches have limits in their ability to adequately synthesize and portray node-related data and nearby connections within the graph structure. Furthermore, they underutilize the potential of heterogeneous network data and fail to adapt or customize it in response to the dynamic nature of learning processes. To address these issues, we propose a new method named AGNE, short for \(\underline{A}ttentional\,\underline{G}raph\,Convolutional\,\underline{N}etwork\,\underline{E}mbedding\), consists of three components. Firstly, we construct a heterogeneous information network that includes knowledge concepts, students, and other entities like videos, courses, and instructors. We employ meta-path embedding techniques to generate node embeddings enriched with semantic content and utilize Graph Convolutional Networks to integrate contextual information of these nodes. Secondly, we utilize a graph attention network to augment the efficacy of information dissemination among the entities. Lastly, we implement matrix factorization to refine and enhance the recommendation algorithm. These components are systematically amalgamated to produce more precise recommendation outcomes. Extensive experiments have been conducted on the large public MOOCCubeX dataset, demonstrating the superiority of AGNE against several state-of-the-art methods.