0405119001 - Sofia Umaroh, S.Pd., M.T Dosen Pembimbing 1 MUHAMMAD HILMY AIMAN / 162021001 Penulis

Penelitian ini dilatarbelakangi oleh kebutuhan untuk mengidentifikasi kolaborator penelitian yang relevan secara lebih akurat, karena metode konvensional yang hanya mengandalkan kesamaan bidang atau riwayat publikasi masih terbatas dalam menangkap kompleksitas hubungan antar peneliti. Untuk mengatasi hal tersebut, penelitian ini mengusulkan sistem rekomendasi berbasis knowledge graph yang dibangun dari ontologi profil peneliti, serta menerapkan Relational Graph Convolutional Network (R-GCN) dengan pendekatan link prediction menggunakan DistMult. Metode penelitian meliputi perancangan ontologi, pembangunan knowledge graph dengan entitas Person, Field of Study, dan Publication, serta relasi antar entitas, kemudian node embedding diperoleh melalui R-GCN dan dipakai untuk memprediksi keterhubungan antar peneliti. Evaluasi dilakukan menggunakan metrik kuantitatif seperti Hits@10, Mean Reciprocal Rank (MRR), dan AUC, serta validasi tambahan melalui penilaian 14 peneliti dengan average relevance score, precision@10, dan NDCG@10. Hasil menunjukkan bahwa model R-GCN tanpa input fitur lebih unggul (Hits@10 = 0,9733; MRR = 0,9422; AUC = 0,7937) dibanding model dengan fitur tambahan, sementara penilaian pengguna juga menegaskan hal ini dengan rata-rata relevance score 3,314 pada model tanpa fitur dan 2,571 pada model dengan fitur. Penelitian ini membuktikan efektivitas knowledge graph dan R-GCN dalam sistem rekomendasi kolaborator, meskipun masih terdapat keterbatasan terkait kualitas data, validasi ontologi, dan generalisasi model, sehingga dapat menjadi landasan bagi pengembangan sistem rekomendasi yang lebih optimal di masa depan. This research is motivated by the need to identify relevant research collaborators more accurately, because conventional methods that only rely on field similarities or publication history are still limited in capturing the complexity of relationships between researchers. To overcome this, this study recommends a knowledge graph-based recommendation system built from researcher profile ontology, and applies Relational Graph Convolutional Network (R-GCN) with a link prediction approach using DistMult. The research method includes ontology design, knowledge graph construction with Person, Field of Study, and Publication entities, as well as relationships between entities, then node embedding is obtained through R-GCN and used to predict the connection between researchers. Evaluation is carried out using quantitative metrics such as Hits@10, Mean Reciprocal Rank (MRR), and AUC, as well as additional validation through the assessment of 14 researchers with an average score of relevance, precision@10, and NDCG@10. The results show that the R-GCN model without feature input is superior (Hits@10 = 0.9733; MRR = 0.9422; AUC = 0.7937) compared to the model with additional features, while user assessments also indicate this with an average relevance score of 3.314 in the model without features and 2.571 in the model with features. This study proves the effectiveness of knowledge graphs and R-GCN in collaborative recommendation systems, although there are still limitations related to data quality, ontology validation, and model generalization, so that it can be a foundation for the development of more optimal recommendation systems in the future.