Interact-RAG: Reason and Interact with the Corpus, Beyond Black-Box Retrieval

2025-10-31

AI Summary

Existing RAG methods treat retrieval as a black-box operation, limiting their capability to handle complex information-seeking tasks. This paper introduces Interact-RAG, the first framework to transform large language model agents from passive query issuers into autonomous decision-makers capable of actively controlling the retrieval process. Its core innovation lies in a reasoning-enhanced workflow and corpus interaction engine, which incorporates programmable interaction primitives—such as *focus*, *backtrack*, and *expand*—to enable fine-grained, dynamic intervention in retrieval paths. Interact-RAG is trained end-to-end via synthetic interaction trajectory generation, supervised fine-tuning, and reinforcement learning. Evaluated on six diverse benchmarks, it substantially outperforms state-of-the-art methods, demonstrating the effectiveness and generalizability of the “reasoning + interaction” co-design paradigm for adaptive, controllable retrieval.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Multiagent Systems: Adversarial Agents
• Natural Language Processing: Question Answering

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Retrieval-Augmented Generation (RAG) has significantly enhanced LLMs by incorporating external information. However, prevailing agentic RAG approaches are constrained by a critical limitation: they treat the retrieval process as a black-box querying operation. This confines agents' actions to query issuing, hindering its ability to tackle complex information-seeking tasks. To address this, we introduce Interact-RAG, a new paradigm that elevates the LLM agent from a passive query issuer into an active manipulator of the retrieval process. We dismantle the black-box with a Corpus Interaction Engine, equipping the agent with a set of action primitives for fine-grained control over information retrieval. To further empower the agent on the entire RAG pipeline, we first develop a reasoning-enhanced workflow, which enables both zero-shot execution and the synthesis of interaction trajectories. We then leverage this synthetic data to train a fully autonomous end-to-end agent via Supervised Fine-Tuning (SFT), followed by refinement with Reinforcement Learning (RL). Extensive experiments across six benchmarks demonstrate that Interact-RAG significantly outperforms other advanced methods, validating the efficacy of our reasoning-interaction strategy.

Problem

• Overcoming black-box limitations in retrieval-augmented generation systems
• Enabling fine-grained agent control over information retrieval processes
• Developing autonomous agents for complex information-seeking tasks

Innovation

• Corpus Interaction Engine enables fine-grained retrieval control
• Reasoning-enhanced workflow supports zero-shot execution and synthesis
• Supervised Fine-Tuning and Reinforcement Learning train autonomous agent