LLM-Based Data Augmentation and MoE Improve Person-Job Fit
A new method for Person-Job Fit (PJF) in online recruitment uses large language model (LLM) data augmentation and a category-aware Mixture of Experts (MoE) module. The LLM-based augmentation polishes low-quality job descriptions via chain-of-thought prompts, while the MoE incorporates category embeddings to distinguish similar candidate-job pairs. Offline and online A/B tests on a recruitment platform show relative improvements of 2.40% in AUC and 7.46% in GAUC, with increased click-through rates. The approach addresses challenges of poor job descriptions and candidate-job similarity.
Key facts
- Method uses LLM-based data augmentation with chain-of-thought prompts.
- Category-aware MoE module assigns dynamic weights to experts.
- Offline and online A/B tests conducted on recruitment platform.
- Relative improvement of 2.40% in AUC and 7.46% in GAUC.
- Boosts click-through rate.
- Addresses low-quality job descriptions and similar candidate-job pairs.
- Paper published on arXiv with ID 2604.21264.
- Proposed method surpasses existing methods.
Entities
Institutions
- arXiv