This repository is a curated literature review focused on foundation models for recommender systems, targeting researchers and practitioners interested in the evolution beyond traditional ID-based methods. It compiles papers exploring the necessity of ID embeddings, the shift towards generative paradigms, and the integration of Large Language Models (LLMs) and multimodal data in recommendation.

How It Works

The collection categorizes papers into key research themes: ID vs. LLM/Multimodal approaches, datasets for transferable/multimodal recommendations, surveys on LLMs in recommendation, and specific sub-areas like parameter-efficient tuning, generative models, and prompt-based recommendation. This structured approach allows users to navigate the landscape of modern recommender system research, highlighting advancements in transferable, multimodal, and LLM-integrated models.

Quick Start & Requirements

This is a literature review, not a software package. No installation or execution is required. The primary resource is the list of papers, many with links to their respective arXiv pages or official code repositories.

Highlighted Details

• Extensive coverage of recent surveys and perspective papers on LLMs and multimodal data in recommendation.
• Lists key datasets (e.g., NineRec, TenRec, PixelRec, MIND) for evaluating transferable and multimodal recommendation models.
• Organizes papers by specific research directions: ID-agnostic models, LLM tuning, generative approaches, and prompt-based methods.
• Includes links to related GitHub repositories and Medium articles for further exploration.

Maintenance & Community

The repository encourages contributions via issues and pull requests, indicating an active community effort to maintain and expand the list. Links to related GitHub profiles are provided.

Licensing & Compatibility

The repository itself does not specify a license. The linked papers are subject to their own respective licenses and publication terms.

Limitations & Caveats

As a literature review, it does not provide executable code or benchmarks. The rapidly evolving nature of LLMs and recommender systems means the list may not be exhaustive or perfectly up-to-date.