Graph Reliability and Integrity Management System (GRIMS)

The U.S. Air Force’s need for real-time situational awareness, threat detection, and pattern-of-life analysis in complex environments highlights the critical role of dynamic knowledge graphs (KGs). KGs enable rich representations of relationships between entities, supporting deep analysis and uncovering hidden patterns. However, keeping these graphs current in rapidly evolving datasets presents a significant challenge. Each update can trigger cascading changes across the KG, demanding either substantial machine learning (ML) intervention or intensive manual oversight to ensure data accuracy. Current ML-driven approaches are limited by their "black box" nature, making their outputs difficult to trust without exhaustive user validation - a bottleneck in mission-critical contexts. To address these issues, we propose the Graph Reliability and Integrity Management System (GRIMS), a solution that balances ML-driven automation with enhanced user control, optimizing KG updates while maintaining data integrity. GRIMS will streamline the KG update process by automating high-confidence updates while flagging medium-confidence updates for user review. In Phase I, we will demonstrate this approach with the BioDDEx and PubGraph knowledge graphs, prioritizing validation for impactful or uncertain updates and allowing user oversight as needed. A key feature of GRIMS is its user-centric interface designed to increase usability and oversight. The interface provides analysts with granular control over the KG update process by presenting update recommendations alongside ML confidence scores. High-confidence updates are applied automatically but displayed for optional user review, ensuring transparency. Medium- and low-confidence updates are flagged with clear visual indicators and contextual information, prompting users to validate or modify these changes. Users will have the ability to revert any automated updates, ensuring they retain ultimate control over graph integrity. Additionally, users can view node properties, trace the origin of updates, and directly edit relationships as needed. This approach combines efficiency with user validation, building trust in the system while reducing manual workload. The Phase I objectives include defining operational requirements, identifying ML models capable of confidence-scored KG updates, and building a user interface that allows streamlined interaction with the graph. By empowering users with actionable insights and transparent control mechanisms, GRIMS will enable the Air Force to maintain accurate, mission-critical knowledge graphs in real time.