A Comprehensive Survey on Retrieval Methods in Recommender Systems (arXiv 2407.21022)

Forum topic · 小凯 · 2026-07-05

Summary

This survey, published on arXiv in July 2024 by Junjie Huang, Jizheng Chen, Jianghao Lin, Jiarui Qin, Ziming Feng, Weinan Zhang and colleagues, examines the retrieval stage of industrial recommender systems—a critical yet often overlooked component of multi-stage cascade ranking pipelines. While most research focuses on ranking, retrieval methods sift through vast candidate pools to filter irrelevant items before ranking prioritizes them. The survey organizes existing work into three key areas: improving similarity computation between users and items, enhancing indexing mechanisms for efficient retrieval, and optimizing retrieval training methods. The authors provide benchmarking experiments on three public datasets and an industrial case study covering the full retrieval process and online serving, including practical challenges. The paper also discusses interfaces with emerging paradigms such as LLM tool use, reinforcement learning, and multi-agent collaboration, and outlines open problems including evaluation credibility, latency and cost constraints, safety, and cross-lingual/multimodal extension. It serves as a foundational reference for researchers and engineers optimizing the retrieval component of cascade recommender systems. Source: https://arxiv.org/abs/2407.21022

A Comprehensive Survey on Retrieval Methods in Recommender Systems

Overview

| Field | Detail | |------|------| | Title | A Comprehensive Survey on Retrieval Methods in Recommender Systems | | Authors | Junjie Huang, Jizheng Chen, Jianghao Lin, Jiarui Qin, Ziming Feng, Weinan Zhang, et al. (7 authors) | | Published | 2024-07-11 | | Link | https://arxiv.org/abs/2407.21022 | | Type | Survey |

One-line Summary

A systematic survey of the retrieval stage in recommender systems, covering similarity computation, indexing, and training optimization, with benchmarks and an industrial case study.

Background and Motivation

In an era of information overload, effective recommender systems are essential. Industry widely uses multi-stage cascade ranking, with retrieval and ranking as the two typical stages: retrieval sifts through vast candidate sets to filter irrelevant items, while ranking prioritizes the survivors. Unlike most studies focused on the ranking stage, this survey targets the critical but often overlooked retrieval stage.

Core Contributions

A unified perspective organizing scattered retrieval research into a comparable framework
Clear decomposition of method components for engineering practice
A comprehensive set of benchmarking experiments on three public datasets
Discussion of interfaces with emerging paradigms (LLM tool use, reinforcement learning, multi-agent collaboration)
Explicit listing of open problems: evaluation credibility, latency and cost, hallucination and safety, cross-lingual and multimodal extension

Three Key Research Areas

1. Improving similarity computation between users and items (e.g., better user/item representation learning) 2. Enhancing indexing mechanisms for efficient large-scale retrieval 3. Optimizing training methods of retrieval models

Industrial Case Study

The survey highlights current industrial applications through a case study on retrieval practices at a specific company, covering the entire retrieval pipeline and online serving, with practical implications and challenges.

Original Abstract (verbatim)

> In an era dominated by information overload, effective recommender systems are essential for managing the deluge of data across digital platforms. Multi-stage cascade ranking systems are widely used in the industry, with retrieval and ranking being two typical stages. Retrieval methods sift through vast candidates to filter out irrelevant items, while ranking methods prioritize these candidates to present the most relevant items to users. Unlike studies focusing on the ranking stage, this survey explores the critical yet often overlooked retrieval stage of recommender systems. To achieve precise and efficient personalized retrieval, we summarize existing work in three key areas: improving similarity computation between user and item, enhancing indexing mechanisms for efficient retrieval, and optimizing training methods of retrieval. We also provide a comprehensive set of benchmarking experiments on three public datasets. Furthermore, we highlight current industrial applications through a case study on retrieval practices at a specific company, covering the entire retrieval process and online serving, along with practical implications and challenges. By detailing the retrieval stage, which is fundamental for effective recommendation, this survey aims to bridge the existing knowledge gap and serve as a cornerstone for researchers interested in optimizing this critical component of cascade recommender systems.

Takeaways for Practitioners

1. Architecture: Cascade retrieval + ranking remains mainstream, but agentic paradigms are making retrieval strategy itself a learnable object. 2. Data: High-quality training signals and click/session logs are as important as model architecture. 3. Evaluation: Offline metrics increasingly diverge from online satisfaction; cross-validate with human evaluation. 4. Deployment: Latency, cost, interpretability, and safety are hard industrial constraints that cannot be ignored in favor of academic benchmarks.

References

Original paper: A Comprehensive Survey on Retrieval Methods in Recommender Systems