HypoAgent Framework for Interactive Hypothesis Generation over Knowledge Graphs
HypoAgent, a novel framework, facilitates interactive abductive hypothesis creation utilizing knowledge graphs. It overcomes the shortcomings of current controllable techniques by allowing for multi-turn dialogues and detailed diagnostics. This framework comprises three distinct agents: the Intent Recognition Agent, which translates user inputs and dialogue history into actionable KG conditions; the Hypothesis Generation Agent, which generates hypotheses based on the identified intentions; and the Root Cause Analysis Agent, which evaluates unreliable hypothesis components through KG neighborhoods. Users can iteratively enhance their hypotheses via natural language communication. This research is available on arXiv with the identifier 2605.31370.
Key facts
- HypoAgent is an agentic framework for interactive abductive hypothesis generation over knowledge graphs.
- It integrates three agents: Intent Recognition, Hypothesis Generation, and Root Cause Analysis.
- The Intent Recognition Agent grounds user utterances and dialogue history into executable KG conditions.
- The Hypothesis Generation Agent performs controllable hypothesis generation according to user intention.
- The Root Cause Analysis Agent diagnoses unreliable hypothesis fragments using KG neighborhood.
- Existing controllable methods struggle with grounding evolving natural-language intents across multi-turn dialogues.
- The framework provides fine-grained diagnosis when generated hypotheses fail.
- The research is published on arXiv with identifier 2605.31370.
Entities
Institutions
- arXiv