Dialograph
A decision-theoretic framework for proactive educational dialogue using temporal graphs.
Dialograph is a lightweight Python library for representing, evolving, and traversing dialogue memory as a temporal graph. Designed for proactive dialogue agents, where reasoning over user preferences, beliefs, and strategies matters more than raw generation.
Core Concepts
- Typed Nodes — user preferences, beliefs/facts, and dialogue strategies — each carrying confidence and temporal metadata.
- Directional, Weighted, Time-Aware Edges — semantic relations (supports, contradicts, elicits), dialogue flow dependencies, and strategy activation paths.
- Decision-Theoretic Reasoning — temporal decay, contradiction detection, and centrality-based scoring for proactive dialogue planning.
Paper in review at the 2026 International Conference on Applied Artificial Intelligence (2AI).
Links: GitHub · Paper (TeX)