# Integrating Large Language Models into Recommendation via Mutual Augmentation and Adaptive Aggregation

**Authors:** Sichun Luo, Yuxuan Yao, Bowei He, Yinya Huang, Aojun Zhou, Xinyi Zhang, Yuanzhang Xiao, Mingjie Zhan, Linqi Song  
**arXiv:** [2401.13870](https://arxiv.org/abs/2401.13870)  
**Date:** 2024-01-25  
**Summary page:** https://www.summarizepaper.com/en/paper/2401.13870v1/

## Key points

- Authors highlight advancements made by conventional recommendation methods through collaborative or sequential data analysis
- Conventional methods excel at mining collaborative information and modeling sequential behavior but struggle with data sparsity and the long-tail problem
- Large Language Models (LLMs) leverage rich textual contexts but face challenges in extracting collaborative or sequential information effectively
- Proposed framework called Llama4Rec synergistically combines conventional recommendation models with LLM-based models
- Llama4Rec introduces data augmentation and prompt augmentation strategies to enhance strengths of each model type
- Adaptive aggregation module refines final recommendation results by combining predictions from both types of models effectively
- Empirical studies on real-world datasets validate superiority of Llama4Rec over baseline methods, showing significant improvements in recommendation performance
- Research contributes towards bridging gap between different recommendation techniques and showcases benefits of integrating diverse methodologies for enhanced recommendations

## Summary

In their paper titled "Integrating Large Language Models into Recommendation via Mutual Augmentation and Adaptive Aggregation," authors Sichun Luo, Yuxuan Yao, Bowei He, Yinya Huang, Aojun Zhou, Xinyi Zhang, Yuanzhang Xiao, Mingjie Zhan, and Linqi Song delve into the realm of recommendation systems. They highlight the significant advancements made by conventional recommendation methods through collaborative or sequential data analysis from user behavior. The authors acknowledge that while conventional recommendation methods excel at mining collaborative information and modeling sequential behavior, they often struggle with issues like data sparsity and the long-tail problem. On the other hand, LLMs are adept at leveraging rich textual contexts but face challenges in extracting collaborative or sequential information effectively. Recognizing these strengths and weaknesses of both approaches, the authors propose a novel framework called Llama4Rec - Large Language Model with Mutual Augmentation and Adaptive Aggregation for Recommendation. Llama4Rec is designed to synergistically combine conventional recommendation models with LLM-based models. The framework introduces data augmentation and prompt augmentation strategies tailored to enhance each model type's strengths. An adaptive aggregation module is also integrated to refine final recommendation results by combining predictions from both types of models effectively. Empirical studies conducted on three real-world datasets validate the superiority of Llama4Rec over baseline methods. The results consistently demonstrate significant improvements in recommendation performance achieved through this innovative approach that leverages the combined potential of conventional methods and LLMs. This research contributes towards bridging the gap between different recommendation techniques and showcases the benefits of integrating diverse methodologies for enhanced recommendations in various domains.