The selection of thesis supervisors is often done manually, which tends to be time-consuming in matching students' research topics with the expertise of faculty members. This study develops a thesis supervisor recommendation system based on the title and abstract of students' final projects, integrating Latent Dirichlet Allocation (LDA), Term Frequency-Inverse Document Frequency (TF-IDF), and Bidirectional Encoder Representations from Transformers (BERT). The research dataset includes 1,096 records from 71 faculty members in the Informatics Engineering Department at Universitas Dian Nuswantoro, collected through Google Scholar. The analysis process begins with text preprocessing such as case folding, tokenization, and stemming, followed by topic analysis using LDA, term-specific weighting through TF-IDF, and context-rich vector representation using BERT. The model matches students' research topics with faculty expertise using Cosine Similarity. Evaluation results show an accuracy of 80%, precision of 66%, and recall of 19%, indicating that the model can provide accurate recommendations, though some relevant items are still missed. This model proves effective in facilitating the selection of thesis supervisors. This research is expected to assist students in finding suitable supervisors and help faculty members identify students with relevant research interests.