A-MapReduce: Executing Wide Search via Agentic MapReduce

2026-02-01

AI Summary

This work addresses the inefficiency of existing large language model–based multi-agent systems in breadth-oriented search tasks, which struggle to meet the demands of large-scale, breadth-first information retrieval. Inspired by the MapReduce paradigm, the study introduces a novel horizontal parallel retrieval mechanism into multi-agent wide search, enabling efficient parallel processing through task-adaptive decomposition and result aggregation. Furthermore, an experience memory module is integrated to dynamically optimize query-driven task allocation and recombination. This approach overcomes the limitations of conventional vertical reasoning frameworks, achieving state-of-the-art performance across five benchmarks with Item F1 improvements of 5.11%–17.50% and a 45.8% reduction in runtime, significantly outperforming existing methods while offering superior cost-effectiveness.

Technology Category

• Multiagent Systems: Multiagent Learning
• Search and Optimization: Learning to Search
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs

Abstract

Contemporary large language model (LLM)-based multi-agent systems exhibit systematic advantages in deep research tasks, which emphasize iterative, vertically structured information seeking. However, when confronted with wide search tasks characterized by large-scale, breadth-oriented retrieval, existing agentic frameworks, primarily designed around sequential, vertically structured reasoning, remain stuck in expansive search objectives and inefficient long-horizon execution. To bridge this gap, we propose A-MapReduce, a MapReduce paradigm-inspired multi-agent execution framework that recasts wide search as a horizontally structured retrieval problem. Concretely, A-MapReduce implements parallel processing of massive retrieval targets through task-adaptive decomposition and structured result aggregation. Meanwhile, it leverages experiential memory to drive the continual evolution of query-conditioned task allocation and recomposition, enabling progressive improvement in large-scale wide-search regimes. Extensive experiments on five agentic benchmarks demonstrate that A-MapReduce is (i) high-performing, achieving state-of-the-art performance on WideSearch and DeepWideSearch, and delivering 5.11% - 17.50% average Item F1 improvements compared with strong baselines with OpenAI o3 or Gemini 2.5 Pro backbones; (ii) cost-effective and efficient, delivering superior cost-performance trade-offs and reducing running time by 45.8\% compared to representative multi-agent baselines. The code is available at https://github.com/mingju-c/AMapReduce.

Problem

• wide search
• multi-agent systems
• large language models
• breadth-oriented retrieval
• long-horizon execution

Innovation

• A-MapReduce
• wide search
• multi-agent systems
• parallel retrieval
• experiential memory