Verifiable by Construction: Evidence-Anchored LLMs for Explainable Fake News Detection
ExplAI 2025 · CEUR Workshop Proceedings 4141, pp. 168-182
Editorial summary
Before classifier training, experts review and refine features extracted from news articles. For each prediction, a language model receives the influential features and their source-text evidence to draft an explanation. Experiments on LIAR and FakeNewsNet measure classification performance and how explanations relate to the highlighted evidence.
Bibliographic reference
Use the publisher’s record for the citation format required by your journal or organisation.
Andrii Shupta, Pavlo Radiuk, Miroslav Kvassay, Piotr Gaj. "Verifiable by Construction: Evidence-Anchored LLMs for Explainable Fake News Detection". ExplAI 2025 · CEUR Workshop Proceedings 4141, pp. 168-182, 2025.
Test a method against your business task.
Describe the decision your system needs to support and the data you have. We assess whether a method fits, define a prototype and agree how to measure its quality, response time and operating cost.