Reproduce “Enhancing Job Recommendation through LLM-based Generative Adversarial Networks” on OpenResearch

About this paper

Yingpeng Du, Di Luo, Rui Yan, Hongzhi Liu, Yang Song, Hengshu Zhu, Jie Zhang

A novel framework, LGIR, enhances job recommendation by integrating Large Language Models (LLMs) with Generative Adversarial Networks (GANs), mitigating LLM fabrication through interactive resume completion and addressing the few-shot problem by aligning generated resume representations. The method significantly outperforms state-of-the-art baselines on real-world datasets, demonstrating improved matching accuracy.

Key insights

• Guiding LLMs with user interaction history is critical for generating more accurate and relevant resume content, effectively reducing the "fabricated generation" issue.
• Generative Adversarial Networks can effectively refine the latent representations of low-quality, LLM-generated resumes for few-shot users, aligning them with high-quality representations and mitigating the data sparsity problem.
• Combining semantic understanding from LLMs with explicit and implicit user interaction data through a hybrid representation learning approach yields superior recommendation performance.

Summary by alphaXiv.