Scholarlink

Sequential recommendation systems struggle with cold start items due to limited interaction data. We show how Large Language Models (LLMs) can reason over warm-but-unseen items to generate synthetic interactions and improve cold start recommendations. At IKEA, we show how this approach can boost cold start performance while preserving warm start performance on room design data, highlighting how LLM reasoning can enhance recommendation systems.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data Augmentation with LLMs for Cold Start Recommendation in E-Commerce

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Natalija Glisovic,
  • Martin Tegner,
  • Danica Kragic

摘要

Sequential recommendation systems struggle with cold start items due to limited interaction data. We show how Large Language Models (LLMs) can reason over warm-but-unseen items to generate synthetic interactions and improve cold start recommendations. At IKEA, we show how this approach can boost cold start performance while preserving warm start performance on room design data, highlighting how LLM reasoning can enhance recommendation systems.