HypoAgent: An Agentic Framework for Interactive Abductive Hypothesis Generation over Knowledge Graphs

HypoAgent is a new AI framework that uses multiple specialized agents to generate logical hypotheses from knowledge graphs through interactive dialogue. The system excels at understanding evolving user intent across multi-turn conversations and diagnosing why generated hypotheses fail, achieving state-of-the-art performance on both commonsense and biomedical knowledge graphs.

HypoAgent addresses a significant gap in abductive reasoning over knowledge graphs by introducing true interactivity to hypothesis generation. Previous systems allowed explicit user guidance but struggled with natural language nuance and multi-turn context—HypoAgent solves this through a three-agent architecture that mimics human problem-solving workflows. The Intent Recognition Agent translates conversational intent into actionable KG conditions, handling the ambiguity inherent in natural language dialogue. The Hypothesis Generation Agent then produces candidates based on these conditions, while the Root Cause Analysis Agent diagnoses failure points and suggests refinements by probing the knowledge graph's neighborhood structure.

The framework represents meaningful progress in making AI systems more interpretable and collaborative. Rather than treating hypothesis generation as a one-shot task, HypoAgent enables iterative refinement through dialogue, allowing users to progressively guide the system toward relevant answers. This mirrors real investigative processes in scientific research and clinical diagnosis.

The demonstration across both commonsense and biomedical domains signals broad applicability. In biomedical contexts, this capability could accelerate hypothesis generation for drug discovery or disease understanding. For knowledge graph applications generally, interactive hypothesis generation enhances utility for researchers, analysts, and domain experts who need explainable, debuggable reasoning chains rather than opaque predictions.

Developers working with knowledge graphs should monitor this framework's adoption, particularly in enterprise settings where interpretability and iterative refinement drive adoption decisions. The open-source release accelerates potential integration into existing knowledge graph pipelines.

• →HypoAgent uses three specialized agents to enable interactive hypothesis generation over knowledge graphs with multi-turn dialogue support
• →The framework diagnoses why hypotheses fail and suggests refinements through knowledge graph neighborhood analysis
• →Performance achieves state-of-the-art semantic similarity across single-turn, multi-turn, and unconditional settings
• →The system successfully bridges natural language intent and executable KG operations, improving usability for domain experts
• →Open-source availability enables rapid adoption in biomedical, commonsense reasoning, and enterprise knowledge graph applications