|
Comparative Analysis of Transformer-Based Models for Extracting Skills from Information Technology Job Posting using Named Entity Recognition |
|---|---|
| รหัสดีโอไอ | |
| Creator | Anucha Ruangsiriwattanakul |
| Title | Comparative Analysis of Transformer-Based Models for Extracting Skills from Information Technology Job Posting using Named Entity Recognition |
| Contributor | Phanuwat Khanja, Chanida Ruangsiriwattanakul, Krit Chaiwannakoop |
| Publisher | Faculty of Informatics, Mahasarakham University |
| Publication Year | 2569 |
| Journal Title | Journal of Applied Informatics and Technology |
| Journal Vol. | 8 |
| Journal No. | 2 |
| Page no. | 262521 |
| Keyword | ALBERTBERTDistilBERT, Named Entity Recognition, Natural Language Processing, RoBERTaSkill Extraction, Transformer-Based Models, XLM-RoBERTa |
| URL Website | https://ph01.tci-thaijo.org/index.php/jait |
| Website title | Journal of Applied Informatics and Technology |
| ISSN | 3088-1803 |
| Abstract | Skill extraction from job postings is critical for labor market analytics andcurriculum design, yet it remains challenging due to unstructured textand diverse terminology. This study conducts a comparative evaluationof five Transformer-based Named Entity Recognition (NER) models—BERT, RoBERTa, DistilBERT, ALBERT, and XLM-RoBERTa—using acurated bilingual dataset of information technology job postings anno-tated with hard and soft skill entities. The methodology involved stan-dardized preprocessing, BIO tagging, and a 70/15/15 train-validation-testsplit, with model fine-tuning carried out using optimized hyperparame-ters. Evaluation metrics included precision, recall, and F1-score, alongsidecomputational efficiency measures such as training time, inference speed,and memory usage. Experimental results indicate that BERT achievedthe highest F1-score (0.90–0.91), while DistilBERT delivered a favorablebalance between accuracy and efficiency, and ALBERT offered moderateperformance with reduced resource demands. In contrast, RoBERTa andXLM-RoBERTa underperformed in this domain-specific context. Thecontributions of this work include establishing a systematic benchmarkfor Transformer-based NER in IT skill extraction, providing insights intothe trade-offs between accuracy and efficiency, and demonstrating prac-tical implications for human resource analytics and curriculum mapping.These findings advance the understanding of model suitability for domain-specific and bilingual skill extraction tasks and lay the foundation forfuture cross-lingual and cross-domain applications |
