Interpretable AI Model Improves Skills-Aware Talent Recommendations
Researchers have unveiled a new model called CF-RL-TOPSIS, designed for skill-focused talent recommendations. This innovative approach combines several elements: a collaborative filtering mechanism that accounts for transitions, a reinforcement-learning style bandit for job categories, and an entropy-weighted TOPSIS component that uses six semantic proxies. They tested the model on two public datasets, JobHop and Karrierewege, using repeated chronological top-5 rankings and paired Wilcoxon tests. On JobHop, the hybrid model achieved an impressive NDCG@5 score of 0.3040 ± 0.0073, outperforming several other methods. Additionally, the chosen fusion coefficients during validation are transparent, addressing the need for interpretability in these systems.
Key facts
- CF-RL-TOPSIS integrates transition-aware collaborative filtering, a reinforcement-style bandit, and TOPSIS.
- Evaluated on JobHop and Karrierewege benchmarks.
- Achieves NDCG@5 = 0.3040 ± 0.0073 on JobHop.
- Outperforms six baseline models including GRU4Rec and SASRec.
- Fusion coefficients are auditable for interpretability.
- Uses six semantic proxies for TOPSIS branch.
- Published on arXiv with ID 2605.24155.
- Focuses on ICT talent-history data.
Entities
Institutions
- arXiv