Buana, Gandhi Surya (2026) Pengembangan Knowledge-Graph Based Labor Market Intelligence Dinamis Melalui Ekstraksi Keterampilan Tanpa Taksonomi dengan Rekayasa Perintah Large Language Model. Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6026221023_Master_Theses.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6026221023_Master_Theses.pdf - Accepted Version Restricted to Repository staff only Download (15MB) | Request a copy |
Abstract
Kebutuhan tenaga kerja ahli terus meningkat sehingga kesenjangan antara kebutuhan dan ketersediaan keahlian tenaga kerja semakin menonjol. Sejumlah negara telah mengembangkan Labor Market Intelligence (LMI) berbasis knowledge graph (KG) untuk mendukung analisis dan pengambilan keputusan bagi pencari kerja, pelatih karier, pemerintah, dan sektor swasta. Namun, penerapan LMI berbasis KG di Indonesia masih menghadapi kendala, terutama karena keberagaman bahasa dan variasi penulisan pada dokumen lowongan pekerjaan, serta keterbatasan pembaruan peta okupansi dan standar kompetensi yang menyebabkan banyak profesi belum memiliki standar kompetensi yang lengkap. Untuk menjawab tantangan tersebut, penelitian ini mengembangkan LMI berbasis KG yang dinamis dengan memanfaatkan LLM untuk mengekstraksi keterampilan dari dokumen CV dan lowongan pekerjaan tanpa bergantung pada referensi taksonomi. Penelitian ini merancang beberapa variasi rantai perintah LLM untuk mengektraksi keterampilan, dengan luaran terstruktur yang membedakan tools dan skills agar definisi keterampilan lebih konsisten pada tahap evaluasi. Performa ekstraksi kemudian dibandingkan dengan metode penelitian terdahulu (SkillNER dan EscoSkillExtractor) menggunakan ground truth yang dinilai oleh ahli, melalui pelabelan semantik dan pengukuran precissions, recall, dan F1. Keterampilan hasil ekstraksi terbaik selanjutnya digunakan untuk pencocokan CV dan lowongan pekerjaan melalui beberapa metode berbasis kemiripan himpunan, meliputi Jaccard, Dice Coefficient, dan coverage, masing-masing pada variasi exact, semantik, serta satu variasi coverage berbobot. Hasil evaluasi peringkat kandidat menggunakan precision@3, precision@5, nDCG@3, dan nDCG@5 menunjukkan bahwa pencocokan semantik secara konsisten lebih efektif dibanding pencocokan exact. Sebagai luaran akhir, seluruh entitas dan relasi diintegrasikan ke dalam prototipe LMI berbasis KG untuk mendukung investigasi visual kecocokan dan kesenjangan keterampilan. Penelitian ini menunjukkan bahwa ekstraksi keterampilan tanpa taksonomi yang divalidasi secara semantik dan pencocokan berbasis makna/semantik dapat menjadi fondasi pengembangan LMI dinamis di Indonesia, sekaligus meningkatkan keterjelasan alasan kecocokan melalui representasi graf.
======================================================================================================================================
The demand for skilled workers continues to grow, making the gap between labor market needs and available workforce competencies increasingly evident. Several countries have developed Knowledge Graph (KG)-based Labor Market Intelligence (LMI) systems to support analysis and decision-making for job seekers, career coaches, government, and the private sector. However, implementing KG-based LMI in Indonesia remains challenging due to language diversity and variation in how skills are written in job vacancy documents, as well as limited updates to occupational maps and competency standards, which leaves many professions without complete competency frameworks. To address these challenges, this study develops a dynamic KG-based LMI approach by leveraging Large Language Models (LLMs) to extract skills from CVs and job vacancy documents without relying on a predefined skills taxonomy. The study designs several LLM prompt-chain variations for skill extraction, producing a structured output that separates tools and skills to ensure consistent skill definitions during evaluation. Extraction performance is then compared against prior methods (SkillNER and EscoSkillExtractor) using expert-labeled ground truth, through semantic labeling and the computation of precision, recall, and F1-score. The best-performing extraction results are subsequently used to match CVs with job vacancies using multiple set-similarity-based methods, including Jaccard, Dice Coefficient, and coverage, evaluated under exact, semantic, and weighted-coverage variants. Candidate ranking evaluation using precision@3, precision@5, nDCG@3, and nDCG@5 shows that semantic matching consistently outperforms exact matching. As the final output, all extracted entities and relationships are integrated into a KG-based LMI prototype to enable visual investigation of skill fit and skill gaps. The findings indicate that taxonomy-free skill extraction validated semantically, combined with meaning-based (semantic) matching, can serve as a foundation for developing a dynamic LMI system in Indonesia while improving the interpretability of matching outcomes through graph-based representations.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Pasar Tenaga Kerja, Large Language Model, Knowledge Graph, Named Entity Recognition Labor Market, Large Language Model, Knowledge Graph, Named Entity Recognition |
| Subjects: | T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T58.62 Decision support systems |
| Divisions: | Faculty of Information and Communication Technology > Information Systems > 59101-(S2) Master Thesis |
| Depositing User: | Gandhi Surya Buana |
| Date Deposited: | 28 Jan 2026 03:29 |
| Last Modified: | 28 Jan 2026 03:29 |
| URI: | http://repository.its.ac.id/id/eprint/130672 |
Actions (login required)
 |
View Item |