With the rapid advancement of Large Language Models (LLMs), an increasing number of researchers are focusing on Generative Recommender Systems (GRSs). Unlike traditional recommendation systems that rely on fixed candidate sets, GRSs leverage generative capabilities, making them more effective in exploring user interests.
Existing LLM-based GRSs primarily utilize Supervised Fine-Tuning (SFT) to enable LLMs to generate candidate items. Additionally, these systems employ similarity-based grounding methods to map the generated results to real-world items. However, SFT-based training is insufficient for LLMs to fully capture the complex interactive behaviors embedded in recommendation scenarios, and similarity-based grounding struggles with the challenges of long-text matching.
To solve the problems, a research team led by Hui XIONG published their new research on 15 January 2026 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.
The research team proposed GIRL (Generative Job Recommendation based on Large Language Models). Specifically, they designed a reward model to evaluate the matching degree between Curriculum Vitae (CVs) and Job Descriptions (JDs). To fine-tune the LLM-based recommender, they introduced a Proximal Policy Optimization (PPO)-based Reinforcement Learning (RL) method. Furthermore, they proposed a model-based grounding method to improve the accuracy of JD grounding.
The proposed method was extensively evaluated on two real-world datasets, and experimental results demonstrate that GIRL outperforms seven baseline methods, achieving superior recommendation effectiveness. Future research directions include exploring more advanced grounding techniques, expanding datasets for better generalization, and optimizing reinforcement learning strategies for enhanced model performance.
Frontiers of Computer Science
Experimental study
Not applicable
Exploiting large language model with reinforcement learning for generative job recommendations
15-Jan-2026