Reproduce “A Survey of Generative Search and Recommendation in the Era of Large Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yongqi Li, Xinyu Lin, Wenjie Wang, Fuli Feng, Liang Pang, Wenjie Li, Liqiang Nie, Xiangnan He and 1 more
This survey paper provides a comprehensive technical review of generative search and recommendation, outlining a unified four-stage framework for understanding methods leveraging large language models. It details diverse approaches to identifier design, training, and inference, and distinguishes key challenges between generative search and recommendation, while also identifying open problems and envisioning content generation as the next paradigm for information seeking.
Key insights
- Representing documents or items as short, generatable textual identifiers is a central challenge, with various approaches offering trade-offs between semantic meaning, distinctiveness, and scalability.
- While both tasks share a generative paradigm, generative search (short queries, low interaction density) and recommendation (long user profiles, high interaction density) present distinct challenges for LLM application.
- Effective generative retrieval often relies on a combination of generative and discriminative training, along with constrained generation techniques, to ensure valid and relevant outputs and mitigate hallucination.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.