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

Researchers introduced HypoAgent, a three-agent framework for interactive abductive hypothesis generation over knowledge graphs.

It combines intent recognition, controllable hypothesis generation, and root-cause analysis to adapt to evolving user input and fix weak hypotheses.

The system achieved state-of-the-art semantic similarity on commonsense and biomedical graph benchmarks.

Results were reported across single-turn, multi-turn, and unconditional settings, highlighting stronger interactive reasoning for explanatory hypotheses.