<![CDATA[The selection of thesis supervisors is a crucial stage that affects the smooth completion of students’ research. The manual process traditionally used often creates difficulties in finding supervisors whose expertise aligns with the research topic, leading to academic inefficiencies. This study aims to design and develop a thesis supervisor recommendation system based on a hybrid TF-IDF and Sentence-BERT (SBERT) approach to improve the accuracy of matching students’ thesis titles with supervisors’ areas of expertise. The dataset used consists of 60 publications from four areas of expertise and 36 thesis titles from DIKE UGM students. The research stages include data collection, aggregation of supervisor publications, text preprocessing, feature extraction, and evaluation using Accuracy@K and Mean Reciprocal Rank (MRR). Experimental results indicate that the combination of stemming and stopword removal provides the best performance in placing relevant supervisors within the top-3 recommendations. The hybrid TF-IDF + SBERT method demonstrates superior performance compared to single methods, achieving Acc@3 of 0.8056, Acc@5 of 0.8611, and MRR of 0.6607, due to its ability to combine lexical information with semantic context. This study shows that a text-based recommendation system can speed up supervisor assignment and improve the match between research topics and supervisors’ expertise.]]>
Copyrights © 2026
