Tools
Tools
Power, Alex (2025) Comparative Evaluation of Resume–Job Matching Using Transformer-Based Semantic Similarity and TF-IDF. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (863kB) | Preview |
Preview |
PDF (Configuration Manual)
Download (893kB) | Preview |
Abstract
This research proposes a candidate-centric system for matching resumes with relevant job descriptions using transformer-based sentence embeddings. Using semantic similarity scoring, this research removes the need for training data and improves relevance in job recommendations. The proposed system uses Sentence-BERT to embed both the given resume and a subset of job descriptions, comparing them using cosine similarity. This will be done using TF-IDF as a baseline for comparison. They will be evaluated using a combination of quantitative and qualitative metrics. Quantitative metrics include nDCG, MRR, Precision@K, Recall@K, Spearman Correlation, Runtime and Memory Usage. The qualitative evaluation consists of feedback based on participant perception of ranked job descriptions gathered from individuals with experience in hiring, HR or related experience. Across all resume domains, SBERT consistently outperformed TF-IDF on ranking accuracy, achieving higher MRR, nDCG, and stronger alignment with human-judged relevance. SBERT also received higher average participant relevance scores in four of five domains. Although TFIDF demonstrated significantly faster runtime, it required considerably more memory, whereas SBERT exhibited slower inference but lower and more stable memory usage. The findings suggest that transformer-based models provide more accurate and semantically meaningful job recommendations without domain-specific fine-tuning, highlighting their suitability for recruitment contexts where contextual understanding is essential.
Actions (login required)
 |
View Item |