RLRF4Rec: Reinforcement Learning from Recsys Feedback for Enhanced Recommendation Reranking

Large Language Models (LLMs) have demonstrated remarkable performance across diverse domains, prompting researchers to explore their potential for use in recommendation systems. Initial attempts have leveraged the exceptional capabilities of LLMs, such as rich knowledge and strong generalization through In-context Learning, which involves phrasing the recommendation task as prompts. Nevertheless, the performance of LLMs in recommendation tasks remains suboptimal due to a substantial disparity between the training tasks for LLMs and recommendation tasks and inadequate recommendation data during pre-training. This paper introduces RLRF4Rec, a novel framework integrating Reinforcement Learning from Recsys Feedback for Enhanced Recommendation Reranking(RLRF4Rec) with LLMs to address these challenges. Specifically, We first have the LLM generate inferred user preferences based on user interaction history, which is then used to augment traditional ID-based sequence recommendation models. Subsequently, we trained a reward model based on knowledge augmentation recommendation models to evaluate the quality of the reasoning knowledge from LLM. We then select the best and worst responses from the N samples to construct a dataset for LLM tuning. Finally, we design a structure alignment strategy with Direct Preference Optimization(DPO). We validate the effectiveness of RLRF4Rec through extensive experiments, demonstrating significant improvements in recommendation re-ranking metrics compared to baselines. This demonstrates that our approach significantly improves the capability of LLMs to respond to instructions within recommender systems.