# Large Language Model Empowered Embedding Generator for Sequential Recommendation

**Authors:** Qidong Liu, Xian Wu, Wanyu Wang, Yejing Wang, Yuanshao Zhu, Xiangyu Zhao, Feng Tian, Yefeng Zheng  
**arXiv:** [2409.19925](https://arxiv.org/abs/2409.19925)  
**Date:** 2024-09-30  
**Summary page:** https://www.summarizepaper.com/en/paper/2409.19925v1/

## Key points

- Authors propose LLMEmb as a technique to enhance Sequential Recommender Systems (SRS) for recommending less popular items
- LLMEmb leverages Large Language Models (LLM) and introduces Supervised Contrastive Fine-Tuning (SCFT)
- Addressing the long-tail problem faced by SRS leads to decreased user discovery and reduced earnings for vendors
- Importance of incorporating collaborative filtering signals into LLM-generated embeddings
- Introduction of Recommendation Adaptation Training (RAT) for this purpose
- Experiments on real-world datasets show that LLMEmb significantly outperforms existing methods in enhancing recommendation accuracy and addressing the long-tail problem effectively

## Summary

In their paper titled "Large Language Model Empowered Embedding Generator for Sequential Recommendation," authors Qidong Liu, Xian Wu, Wanyu Wang, Yejing Wang, Yuanshao Zhu, Xiangyu Zhao, Feng Tian, and Yefeng Zheng propose LLMEmb as a novel technique to enhance the performance of Sequential Recommender Systems (SRS) in recommending less popular items. They address the long-tail problem faced by SRS which leads to decreased user discovery and reduced earnings for vendors. LLMEmb leverages Large Language Models (LLM) and introduces Supervised Contrastive Fine-Tuning (SCFT) to align general-purpose LLM capabilities with recommendation domain requirements. The authors emphasize the importance of incorporating collaborative filtering signals into LLM-generated embeddings and introduce Recommendation Adaptation Training (RAT) for this purpose. Their experiments on real-world datasets show that LLMEmb significantly outperforms existing methods in enhancing recommendation accuracy and addressing the long-tail problem effectively. This research offers valuable insights into leveraging Large Language Models for sequential recommendation systems and provides a promising solution to improve user experience and vendor earnings in recommendation scenarios.