Pembangunan Sistem Rekomendasi Dosen Pembimbing menggunakan Data Talent Mapping dan Content-Based Filtering

Abstract

Penentuan topik skripsi merupakan tantangan bagi mahasiswa tingkat akhir akibat keterbatasan informasi dan minimnya arahan dari dosen pembimbing. Penelitian ini membangun sistem rekomendasi dosen pembimbing berbasis Content-Based Filtering dengan menggabungkan nilai akademik dan hasil pemetaan bakat sebagai representasi profil mahasiswa. Kata kunci diekstraksi dari silabus mata kuliah dan dokumen skripsi menggunakan metode TF-IDF berbasis ontologi, kemudian kemiripan antara profil mahasiswa dan dokumen skripsi dihitung menggunakan pendekatan cosine similarity dan Distance Extended Wu and Palmer (DEWP). Evaluasi sistem dilakukan melalui dua pendekatan, pertama menggunakan metrik precision, recall, dan F1-score, kedua menggunakan metrik relevance, diversity, serendipity, dan novelty. Hasil evaluasi menunjukkan bahwa kombinasi penggabungan perhitungan menggunakan weighted combination dengan bobot cosine similarity 0,8 dan DEWP 0,2 menghasilkan F1-score terbaik sebesar 37,76%. Namun, nilai precision dan recall yang masih di bawah 40% mengindikasikan keterbatasan sistem dalam merekomendasikan dosen yang sesuai dengan ekspektasi personal mahasiswa. Sebaliknya, evaluasi berbasis persepsi mahasiswa menunjukkan skor relevance 73%, diversity 89%, serendipity 79%, dan novelty 86%, yang menandakan bahwa sistem dianggap mampu memberi saran yang beragam, mengejutkan, dan masih relevan secara konseptual

Determining a thesis topic is a challenge for final-year students due to limited information and minimal guidance from supervisors. This research builds a thesis supervisor recommendation system based on Content-Based Filtering by combining academic scores and talent mapping results as a representation of student profiles. Keywords are extracted from course syllabi and thesis documents using the ontology-based TF-IDF method, and then the similarity between student profiles and thesis documents is calculated using the cosine similarity and Distance Extended Wu and Palmer (DEWP) approaches. The system evaluation was conducted through two approaches: first, using the precision, recall, and F1-score metrics; second, using the relevance, diversity, serendipity, and novelty metrics. The evaluation results show that the combination of calculations using a weighted combination with cosine similarity weight of 0,8 and DEWP 0,2 yields the best F1-score of 37,76%. However, the precision and recall values still below 40% indicate the system's limitations in recommending professors that align with students' personal expectations. On the other hand, student perception-based evaluations show relevance scores of 73%, diversity 89%, serendipity 79%, and novelty 86%, indicating that the system is considered capable of providing diverse, surprising, and conceptually relevant suggestions.