Research Projects

Our research addresses strategic interventions in dynamic opinion networks through complementary approaches. We developed a visualization framework with the first numeric PDDL for opinion dynamics, enabling network simulation and intervention planning. To tackle computational challenges in larger networks, we introduced learning-based strategies combining ranking algorithms with neural network classifiers. Our reinforcement learning framework demonstrated that GCN-based planners achieve effective control across various network configurations, particularly in mitigating misinformation spread. The results show improved performance with increased action budgets and reveal that reward strategies focusing on susceptible nodes and infection rates outperform rapid blocking approaches.

Ontologies effectively organize rich metadata, enable semantic queries for novel insights, and promote reuse. This paper addresses automated planning, which involves finding action sequences to transition from an initial to a goal state. We hypothesize that the wealth of planners and diverse domains contain valuable information for identifying suitable planners and enhancing performance. Using data from the International Planning Competition (IPC), we construct a planning ontology and demonstrate its utility in two use cases: selecting promising planners and improving performance with macros—action ordering constraints derived from the ontology. The ontology and resources are shared to foster further research.