A Survey of Generative Search and Recommendation in the Era of Large Language Models

Forum topic · 小凯 · 2026-07-05

Summary

This April 2024 arXiv survey (arXiv:2404.16924) by Yongqi Li, Xinyu Lin, Wenjie Wang, Fuli Feng, Liang Pang, Wenjie Li and colleagues systematically reviews how large language models (LLMs) are driving a paradigm shift in information access: from traditional ranking-based search and recommendation toward generative approaches. The paper defines generative information retrieval (GenIR) and generative recommendation, where candidate documents or items are directly generated—typically as identifiers or free-form text—rather than scored and ranked by an index. It provides a taxonomy of methods across modeling paradigms, LLM integration patterns (RAG, agents, tool use), optimization objectives, and evaluation protocols, covering dense retrieval, late interaction (e.g., ColBERT), generative retrieval (DSI-style docid generation), and agentic search. The survey also discusses benchmarks (MS MARCO, BEIR, Natural Questions), metrics (nDCG, MRR, Recall@k), open challenges such as hallucination, latency and cost, evaluation reliability, and cross-lingual/multimodal extension, and outlines future directions including retrieval–reasoning co-training and unified benchmarks. It is a useful reference for researchers and engineers building LLM-powered search and recommender systems.

A Survey of Generative Search and Recommendation in the Era of Large Language Models

Source: arXiv:2404.16924 · Type: Survey · Authors: Yongqi Li, Xinyu Lin, Wenjie Wang, Fuli Feng, Liang Pang, Wenjie Li, et al. (9 authors) · Date: April 2024

One-sentence summary

This survey systematizes the shift from traditional rank-based search and recommendation systems to generative approaches powered by large language models (LLMs), providing a unified taxonomy, representative methods, evaluation practices, and open research problems.

Background and motivation

Traditional search and recommendation pipelines decouple retrieval, ranking, and generation, which struggles to meet the LLM-era demands of natural-language interaction, multi-hop reasoning, and fresh knowledge. The survey addresses how the field is moving from a rank-based paradigm (indexing documents/items and scoring them) to a generative paradigm, where the model directly generates the required information—document identifiers, item IDs, or natural-language answers/recommendations.

The core scenarios covered include open-domain information access, conversational search, semantic understanding in recommender systems, and end-to-end architectures that combine external knowledge sources with generative models.

Core contributions

A unified perspective that brings scattered generative search and recommendation work into a comparable framework.
A clear decomposition of method components: representation learning, retrievers, re-rankers, planners, generators, and feedback mechanisms.
A structured taxonomy and coverage of benchmarks/datasets that lowers the entry barrier for follow-up research.
Discussion of interfaces with emerging paradigms such as LLM tool use, reinforcement learning, and multi-agent collaboration.
Explicitly listed open problems: evaluation reliability, latency and cost, hallucination and safety, cross-lingual and multimodal extension.

Taxonomy of methods

| Dimension | Sub-categories | Representative ideas | Strengths | Limitations | |---|---|---|---|---| | Modeling paradigm | Discriminative retrieval / Generative retrieval | Bi-encoders, cross-encoders, DSI, GPT-style indexing | Mature, scalable | Semantic drift, index update cost | | LLM integration | RAG / Agent / Tool use | Retrieval augmentation, search agents, API calls | Flexible, interpretable | Latency, error propagation | | Optimization objectives | Relevance / diversity / freshness | Multi-objective LTR, RLHF, online learning | Business-aligned | Annotation scarcity | | Evaluation | Offline / Online / Human | nDCG, MRR, LLM-as-judge, A/B tests | Comparable | Deviates from true satisfaction |

Main research lines compared

The survey situates four major lines of work:

Dense retrieval — high recall, low latency; suited to first-stage retrieval.
Late interaction (e.g., ColBERT) — higher accuracy but larger index footprint.
Generative IR — directly "generates" documents via tokens or docids, simplifying the cascade.
Agentic search — models search as sequential decision-making, supporting multi-hop reasoning and self-reflection.

Timeline and evolution

2019–2021: BERT re-ranking and DPR establish neural retrieval foundations.
2022–2023: RAG and FreshLLM drive retrieval–generation fusion.
2024 onward: Conversational/agentic search and generative recommender systems (Gen-RecSys) flourish.
Emerging: RL-trained search agents, deep research systems, and GraphRAG.

Evaluation paradigms

Typical setups covered include:

Datasets: MS MARCO, BEIR, Natural Questions, domain corpora, public recommendation datasets.
Metrics: nDCG@10, MRR, Recall@k, Hit@k, human preference, task success rate, latency and token cost.
Baselines: BM25, dense retrieval, cross-encoder re-ranking, retrieval-free LLMs, commercial search APIs.
Ablations: contribution of retrieval steps, re-ranking depth, and training data scale.

Consult the original PDF for exact quantitative tables.

Key insights

1. Architecture: cascaded retrieve → re-rank → generate remains mainstream, but agentic paradigms make "when and how much to retrieve" itself learnable. 2. Data: high-quality instruction data and click/session logs are both critical; synthetic data risks knowledge leakage and distribution shift. 3. Evaluation: the gap between offline metrics and online satisfaction is widening; LLM-as-judge needs cross-validation with human evaluation. 4. Products: latency, cost, explainability, and safety are hard constraints for industrial deployment.

Open problems and future directions

The authors highlight: lack of unified benchmarks, non-reproducibility of private data, LLM evaluation bias, and safety/cost constraints of agentic systems. Future work includes finer-grained process supervision, joint retrieval–reasoning training, enterprise metadata governance, and multimodal/cross-lingual consistency, as well as deeper fusion with knowledge graphs and structured databases, and causal/fairness constraints for recommendation.

Practical engineering checklist

| Item | Question | Suggestion | |---|---|---| | Data | Does training/indexing contain PII? Version control? | Partitioned indexes, anonymization, rollback-capable embedding versions | | Latency | p99 budget? How many retrieval steps? | Cascade + early exit, caching hot queries, async re-ranking | | Quality | Do offline gains translate to online CTR/satisfaction? | Interleaving experiments, human audits, citation verification | | Safety | Does open retrieval introduce poisoning/bias? | Source whitelists, adversarial detection, output filtering | | Cost | Token and GPU cost per query? | Small-model routing, distillation, hybrid sparse+dense retrieval |

Glossary

IR — Information Retrieval
RAG — Retrieval-Augmented Generation
LTR — Learning to Rank
nDCG — Normalized Discounted Cumulative Gain, a ranking quality metric
Agentic Search — modeling search as sequential decision-making with tool calls
Gen-IR — Generative Information Retrieval

References

Original paper: *A Survey of Generative Search and Recommendation in the Era of Large Language Models*. https://arxiv.org/abs/2404.16924

A Survey of Generative Search and Recommendation in the Era of Large Language Models

Summary

A Survey of Generative Search and Recommendation in the Era of Large Language Models

One-sentence summary

Background and motivation

Core contributions

Taxonomy of methods

Main research lines compared

Timeline and evolution

Evaluation paradigms

Key insights

Open problems and future directions

Practical engineering checklist

Glossary

Related reading

References

Tags