A Survey on Large Language Models for Recommendation (WWW 2024, Springer)
This entry indexes a survey published at WWW 2024 in the Springer *World Wide Web* journal, examining the intersection of large language models (LLMs) and recommender systems.
Metadata
| Field | Content | |-------|---------| | Type | Survey | | Venue | WWW 2024, Springer | | Link | https://link.springer.com/article/10.1007/s11280-024-01291-2 | | Section | Recommender Engine Surveys |
Scope and Motivation
Traditional recommender pipelines separate retrieval, ranking, and generation, which struggles to meet modern demands for natural-language interaction, multi-hop reasoning, and up-to-date knowledge. This survey systematically reviews how LLMs reshape recommendation, providing a unified perspective on a rapidly growing body of work.
Key Themes Covered
- Unified taxonomy of LLM-for-recommendation approaches, making scattered prior work comparable.
- Component decomposition: representation learning, retrievers, re-rankers, planners, generators, and feedback mechanisms.
- Integration paradigms:
- *RAG / retrieval augmentation* — grounding LLM outputs in external item/user knowledge;
- *Agentic / tool-use* — modeling recommendation as sequential decision-making with API calls;
- *Generative recommendation (Gen-Rec)* — directly generating items or identifiers.
- Optimization objectives: relevance, diversity, freshness; multi-objective learning-to-rank, RLHF, online learning.
- Evaluation: offline metrics (nDCG, MRR, Recall@k, Hit@k), online A/B testing, human preference studies, and emerging LLM-as-judge protocols.
- 2019–2021: BERT re-ranking and dense retrieval (DPR) establish neural foundations.
- 2022–2023: RAG-style retrieval–generation fusion matures.
- 2024 onward: conversational/agentic search and generative recommender systems (Gen-RecSys) expand rapidly.
- 2025+: RL-trained search agents, Deep Research, and GraphRAG emerge as growth areas.
- Architecture: cascade retrieve → re-rank → generate remains mainstream, with agentic methods making retrieval policy itself learnable.
- Data: high-quality instruction data and interaction logs matter; synthetic data risks leakage and distribution shift.
- Deployment: latency, cost, explainability, and safety are hard engineering constraints — do not optimize academic benchmarks alone.
- A Comprehensive Survey on Retrieval Methods in Recommender Systems
- A Survey of Generative Search and Recommendation in the Era of Large Language Models
- A Survey on LLM-powered Agents for Recommender Systems
- A Comprehensive Survey on Cross-Domain Recommendation
Research Timeline Context
Open Challenges
1. Lack of unified benchmarks; private data limits reproducibility. 2. LLM evaluation bias; growing gap between offline metrics and real user satisfaction. 3. Latency, token cost, hallucination, and safety constraints in production deployment. 4. Cross-lingual and multimodal generalization, especially given English-centric training data.