Enhancing Skills Demand Understanding through Job Ad Segmentation Using NLP and Clustering Techniques

Abstract

The labor market has been significantly impacted by the rapidly evolving global landscape, characterized by increased competition, globalization, demographic shifts, and digitization, leading to a demand for new skills and professions. The rapid pace of technological advancements, economic transformations, and changes in workplace practices necessitate that employees continuously adapt to new skill requirements. A quick assessment of these changes enables the identification of skill profiles and the activities of economic fields. This paper aims to utilize natural language processing technologies and data clustering methods to analyze the skill needs of Lithuanian employees, perform a cluster analysis of these skills, and create automated job profiles. The hypothesis that applying natural language processing and clustering in job profile analyzes can allow the real-time assessment of job skill demand changes was investigated. Over five hundred thousand job postings were analyzed to build job/position profiles for further decision-making. In the first stage, data were extracted from the job requirements of entire job advertisement texts. The regex procedure was found to have demonstrated the best results. Data vectorization for initial feature extraction was performed using BERT structure transformers (sentence transformers). Five dimensionality reduction methods were compared, with the UMAP technique producing the best results. The HDBSCAN method proved to be the most effective for clustering, though RCBMIDE also demonstrated a robust performance. Finally, job profile descriptions were generated using generative artificial intelligence based on the compiled job profile skills. Upon expert assessment of the created job profiles and their descriptions, it was concluded that the automated job advertisement analysis algorithm had shown successful results and could therefore be applied in practice.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 29 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.