Explainable AI Components for Narrative Map Extraction

Multi-level explanations build appropriate user trust: The study found that providing explanations at different levels of abstraction—from low-level document relationships to high-level narrative patterns—helped users develop appropriate confidence in the AI system's narrative extraction capabilities, rather than either over-trusting or under-trusting it.

Connection explanations and event detection are most effective: Among the explanation types tested, connection explanations (showing how events relate to each other) and important event detection were particularly effective at building user confidence, suggesting these should be prioritized when designing explainable narrative systems.

User studies validate XAI effectiveness: The paper demonstrates the importance of empirical evaluation through user studies (10 participants analyzing 2021 Cuban protest narratives) rather than relying solely on technical metrics, showing that real users found the explanations meaningful and trust-building.

Transparency supports human-AI collaboration: By making narrative extraction systems interpretable and explainable, the research shows that AI can better support human decision-making rather than operating as a "black box," enabling users to understand and verify the system's reasoning about complex information.