Research map / Time series / Long-term forecasting
A Cross-View Hierarchical Graph Learning Hypernetwork for Skill Demand-Supply Joint Prediction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes CHGH, an encoder-decoder graph model with a cross-view graph encoder, hierarchical cluster encoder, and conditional hyper-decoder that jointly predicts skill demand and supply shares from job postings and work experience data.
Problem
Predicting future skill demand and supply jointly in the labor market, where prior work either relies on expert knowledge or treats skill evolution as a simple univariate time series forecasting problem.
Approach
Combines a cross-view graph encoder capturing asymmetric demand-supply skill relations through a learned adaptive matrix, a DiffPool-based hierarchical encoder for cluster-level co-evolution, and a hypernetwork-generated decoder conditioned on historical skill gaps to jointly output demand and supply trend classes.
Key results
- On the IT dataset, CHGH reaches ACC 0.6777, F1 0.6784, AUC 0.8968, J-ACC 0.4674, outperforming the best baseline Wavenet (ACC 0.6170).
- On FIN, CHGH achieves ACC 0.7227 and J-ACC 0.5390 versus MTGNN's ACC 0.6621 and J-ACC 0.4302.
- On CONS, CHGH achieves ACC 0.7704 and J-ACC 0.6236 versus Wavenet's ACC 0.6702 and J-ACC 0.4491.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add a separate adaptive graph learner to the static-graph model (ablation) | Skill demand-supply prediction · IT, FIN, CONS | ACC: IT: 0.5398 -> 0.5724; FIN: 0.4546 -> 0.5826; CONS: 0.5183 -> 0.5660 | Helped |
| Add the cross-view graph encoder (ablation) | Skill demand-supply prediction · IT, FIN, CONS | ACC: IT: 0.5724 -> 0.6179; FIN: 0.5826 -> 0.6839; CONS: 0.5660 -> 0.7167 | Helped |
| Add the hierarchical graph encoder (ablation) | Skill demand-supply prediction · IT, FIN, CONS | ACC: IT: 0.6179 -> 0.6596; FIN: 0.6839 -> 0.7218; CONS: 0.7167 -> 0.7440 | Helped |
| Add the gap-conditioned hyper-decoder (ablation) | Skill demand-supply prediction · IT, FIN, CONS | ACC: IT: 0.6596 -> 0.6777; FIN: 0.7218 -> 0.7227; CONS: 0.7440 -> 0.7704 | Helped |
| Add the gap-conditioned hyper-decoder (ablation) | Skill demand-supply prediction · IT | AUC: 0.9082 -> 0.8968 | Hurt |
| Use CHGH instead of the strongest baseline for each metric | Skill demand-supply prediction · IT, FIN, CONS | ACC, F1, AUC, J-ACC: IT ACC: 0.6170 -> 0.6777; FIN ACC: 0.6621 -> 0.7227; CONS ACC: 0.6702 -> 0.7704 | Helped |
Surprising findings
- Recurrent models (LSTM) outperformed Transformer-based models on these skill datasets, which the authors attribute to Transformers focusing on long-horizon auto-correlation while overlooking trend.
Limitations
- Task is framed as five-class trend classification after discretization rather than regression of exact shares.
- Evaluated on only three datasets from one online recruitment platform over roughly 1.5 years.
Experimental setup
Datasets: online recruitment platform job postings and work experience (IT, FIN, CONS)
Benchmarks: IT dataset FIN dataset CONS dataset
Metrics: ACC weighted F1 AUC J-ACC
Baselines: ARIMA VAR LSTM Transformer Autoformer Wavenet MTGNN
Builds on: DiffPool hypernetwork MTGNN Graph WaveNet
Closest papers by meaning
- Job-SDF: A Multi-Granularity Dataset for Job Skill Demand Forecasting and Benchmarking (NeurIPS 2024 · similarity 0.87)
- PLD: A Choice-Theoretic List-Wise Knowledge Distillation (NeurIPS 2025 · similarity 0.82)
- IOP: An Idempotent-Like Optimization Method on the Pareto Front of Hypernetwork (AAAI 2025 · similarity 0.82)
- HGCN2SP: Hierarchical Graph Convolutional Network for Two-Stage Stochastic Programming (ICML 2024 · similarity 0.81)
- EduLLM: Leveraging Large Language Models and Framelet-Based Signed Hypergraph Neural Networks for Student Performance Prediction (ICML 2025 · similarity 0.81)
- GCF: Estimating Unobserved Demand Using Graph Causal Forecasting (AAAI 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.