Research map / Large language models / AI and society
Your Career Path Matters in Person-Job Fit
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 WEPJM, a person-job fit model that uses career path (work experience sequence) via three auxiliary objectives (career path identification, reconstruction, and job-career contrastive learning) on top of a BERT+BiLSTM hierarchical encoder.
Problem
Person-job fit systems typically ignore the career path section of resumes, which reveals hidden preferences about consistency, likeness, and continuity of a job seeker's work experiences.
Approach
Model the career path as a chronological sequence of work experiences with hierarchical BERT+BiLSTM encoders, and add three auxiliary objectives: identifying replaced work experiences (consistency), detecting shuffled career sequences (continuity), and contrastively aligning the last work experience with the job posting (likeness).
Key results
- WEPJM achieves 0.850 accuracy, 0.837 F1 and 0.929 AUC, outperforming the best baseline PJFCANN (0.813 acc, 0.816 F1, 0.894 AUC).
- Career path identification gives the largest single auxiliary gain (+CPI F1 0.828 vs No Auxiliary 0.803).
- Performance degrades for job seekers with fewer than 3 work experiences, but remains above text-matching baselines.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use WEPJM instead of PJFCANN | Person-job matching · Person-job matching test set | accuracy: 0.813 -> 0.850 | Helped |
| Add career path identification (CPI) (ablation) | WEPJM · Person-job matching test set | accuracy: 0.813 -> 0.827 | Helped |
| Add career path reconstruction (CPR) (ablation) | WEPJM · Person-job matching test set | accuracy: 0.813 -> 0.819 | Helped |
| Add job-career contrastive learning (CL) (ablation) | WEPJM · Person-job matching test set | accuracy: 0.813 -> 0.821 | Helped |
| Add all three auxiliary tasks: CPI, CPR, and CL (ablation) | WEPJM · Person-job matching test set | accuracy: 0.813 -> 0.850 | Helped |
| Add CPI (ablation) | WEPJM with all three auxiliary tasks · Person-job matching test set | accuracy: 0.831 -> 0.850 | Helped |
| Add CPR (ablation) | WEPJM with all three auxiliary tasks · Person-job matching test set | accuracy: 0.842 -> 0.850 | Helped |
| Add CL (ablation) | WEPJM with all three auxiliary tasks · Person-job matching test set | accuracy: 0.834 -> 0.850 | Helped |
Limitations
- Evaluated only on a single proprietary dataset from one platform.
- Performance drops for job seekers with few work experiences.
- Dataset is not released, limiting reproducibility.
Experimental setup
Datasets: in-house online recruitment platform dataset
Benchmarks: in-house online recruitment platform test set
Metrics: accuracy precision recall F1 AUC
Baselines: LR DT NB RF GBDT PJFNN APJFNN JRMPM ResumeGAN PJFCANN BERT BERT+TAPT
Builds on: BERT multi-task learning contrastive learning
Closest papers by meaning
- A Mathematical Framework for AI-Human Integration in Work (ICML 2025 · similarity 0.80)
- Job-SDF: A Multi-Granularity Dataset for Job Skill Demand Forecasting and Benchmarking (NeurIPS 2024 · similarity 0.80)
- Two Tickets are Better than One: Fair and Accurate Hiring Under Strategic LLM Manipulations (ICML 2025 · similarity 0.80)
- JobBench: Aligning Agent Work With Human Will (NeurIPS 2026 · similarity 0.78)
- Probing Social Bias in Labor Market Text Generation by ChatGPT: A Masked Language Model Approach (NeurIPS 2024 · similarity 0.78)
- FairJob: A Real-World Dataset for Fairness in Online Systems (NeurIPS 2024 · similarity 0.78)
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.