Authors: Xiaohan Yu, Li Zhang, Xin Zhao, Yue Wang, Zhongrui Ma

Published on: February 07, 2024

Impact Score: 8.15

Arxiv code: Arxiv:2402.04527

Summary

• What is new: Introduction of a new paradigm, ID representation, which uses pre-trained ID embeddings with LLMs for improved recommendations.
• Why this is important: Existing LLM-based recommendation systems lack recommendation knowledge and uniqueness.
• What the research proposes: RA-Rec, an ID representation alignment framework that incorporates pre-trained ID embeddings as soft prompts and includes an innovative alignment module.
• Results: RA-Rec achieves up to 3.0% absolute HitRate@100 improvements with less than 10x training data compared to current state-of-the-art methods.

Technical Details

Technological frameworks used: RA-Rec

Models used: Multiple ID-based methods and LLM architectures

Data used: Less than 10x training data for alignment

Potential Impact

Digital advertising, e-commerce platforms, content streaming services

Want to implement this idea in a business?

We have generated a startup concept here: RecAlign.