Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Skill Demand Forecasting Using Temporal Knowledge Graph Embeddings

Yousra Fettach, Adil Bahaj, Mounir Ghogho

TL;DR

This work reframes skill demand forecasting as a temporal knowledge graph (TKG) completion task, constructing JobEdKG from Moroccan job ads and MOOC data to model high-order relations among jobs, skills, sectors, and companies. It trains and evaluates multiple temporal KG embedding approaches (including TA and DE families) for temporal link prediction and demonstrates predictive insights for IT skills over time. The study shows TA-based models, particularly TA-DistMult, offer robust performance on JobEdKG, and provides a scalable implementation (TempTorckKGE) with case studies illustrating inferred skill trajectories. The results suggest temporal KG forecasting can offer timely, semantically grounded forecasts to inform education, policy, and employer reskilling decisions in a rapidly evolving labor market.

Abstract

Rapid technological advancements pose a significant threat to a large portion of the global workforce, potentially leaving them behind. In today's economy, there is a stark contrast between the high demand for skilled labour and the limited employment opportunities available to those who are not adequately prepared for the digital economy. To address this critical juncture and gain a deeper and more rapid understanding of labour market dynamics, in this paper, we approach the problem of skill need forecasting as a knowledge graph (KG) completion task, specifically, temporal link prediction. We introduce our novel temporal KG constructed from online job advertisements. We then train and evaluate different temporal KG embeddings for temporal link prediction. Finally, we present predictions of demand for a selection of skills practiced by workers in the information technology industry. The code and the data are available on our GitHub repository https://github.com/team611/JobEd.

Skill Demand Forecasting Using Temporal Knowledge Graph Embeddings

TL;DR

This work reframes skill demand forecasting as a temporal knowledge graph (TKG) completion task, constructing JobEdKG from Moroccan job ads and MOOC data to model high-order relations among jobs, skills, sectors, and companies. It trains and evaluates multiple temporal KG embedding approaches (including TA and DE families) for temporal link prediction and demonstrates predictive insights for IT skills over time. The study shows TA-based models, particularly TA-DistMult, offer robust performance on JobEdKG, and provides a scalable implementation (TempTorckKGE) with case studies illustrating inferred skill trajectories. The results suggest temporal KG forecasting can offer timely, semantically grounded forecasts to inform education, policy, and employer reskilling decisions in a rapidly evolving labor market.

Abstract

Rapid technological advancements pose a significant threat to a large portion of the global workforce, potentially leaving them behind. In today's economy, there is a stark contrast between the high demand for skilled labour and the limited employment opportunities available to those who are not adequately prepared for the digital economy. To address this critical juncture and gain a deeper and more rapid understanding of labour market dynamics, in this paper, we approach the problem of skill need forecasting as a knowledge graph (KG) completion task, specifically, temporal link prediction. We introduce our novel temporal KG constructed from online job advertisements. We then train and evaluate different temporal KG embeddings for temporal link prediction. Finally, we present predictions of demand for a selection of skills practiced by workers in the information technology industry. The code and the data are available on our GitHub repository https://github.com/team611/JobEd.

Paper Structure

This paper contains 25 sections, 10 equations, 10 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background and Related work
  3. Skill Demand Forecasting
  4. Knowledge graphs in analyzing the job market
  5. Temporal KGs embedding
  6. Time series forecasting vs. TKG link forecasting
  7. Methodology
  8. Data: JobEdKG
  9. Temporal KG Link Prediction
  10. Temporal KG.
  11. The task of temporal KG forecasting.
  12. Embedding layer
  13. Scoring function
  14. Training objective
  15. Evaluation metrics
  16. ...and 10 more sections

Figures (10)

  • Figure 1: Overview of the methodology for building T-JobEdKG
  • Figure 2: Metamodel of T-JobEdKG. DD/MM/YYYY stands for the timestamps associated with links (day, month, year)
  • Figure 3: Training and inference on temporal KGs. The model is trained on triples with timestamps. It learns temporal embeddings to rank correct triples higher than incorrect ones. After training, it can predict missing or future relations.
  • Figure 4: Temporal evolution of the "data analysis" skill for the job "Consulting Engineer in IT Management". The red square showcases the prediction of our model for the time range between 01-01-2023 and 01-01-2025.
  • Figure 5: Temporal evolution of the "Java" skill for the job "Consulting Engineer in IT Management". The red square showcases the prediction of our model for the time range between 01-01-2023 and 01-01-2025
  • ...and 5 more figures