Sistem Rekomendasi Mata Kuliah Pilihan Menggunakan Graph Neural Network Pada Knowledge Graph Data Akademik Mahasiswa (Studi Kasus: Fakultas Ilmu Komputer Universitas Jember)

Abstract

The academic planning process is a crucial phase for students at the Faculty of Computer Science, Universitas Jember. However, many students face difficulties in selecting elective courses that align with their capabilities and the faculty's recommended learning paths, often leading to suboptimal study durations. This research aims to develop an intelligent course recommendation system utilizing a Graph Neural Network (GNN) on a Knowledge Graph (KG) to address these issues. The methodology involves constructing a Heterogeneous Knowledge Graph derived from two primary faculty datasets: the official curriculum and historical academic transcripts of students from the 2020-2022 cohorts. The extracted graph structurally models students, courses, explicit prerequisite rules, and academic performances. A Graph Attention Network (GAT) architecture was implemented to perform link regression tasks, predicting the relevance score of un-taken courses based on multi-hop dependencies and self-attention mechanisms. The model evaluation demonstrated outstanding performance with an RMSE of 0.5487, Precision of 0.9494, Recall of 0.9981, F1-Score of 0.9732, MAP of 0.9715, and NDCG of 0.9939. Furthermore, an ablation study proved that the GAT significantly outperformed conventional aggregation algorithms by successfully assigning higher attention weights to relevant prerequisite courses. The study concludes that integrating prerequisite structures into the Knowledge Graph acts as a logical pacing mechanism, accurately guiding students along realistic learning paths. The resulting recommendation model was successfully integrated into a mobile application prototype, allowing students to filter course recommendations based on their specific career interests.