From LLMs to Information Cascades
This presentation covers the full pipeline from LLM-based text annotation to media cascade detection. It introduces LLM Tool, a local-first open-source framework for structured annotation using large language models, and demonstrates its application within the Canadian Climate Framing (CCF) project, which analyzes 47 years of Canadian press coverage on climate change.
March 16, 2026A 100% local, open-source framework for structured text annotation. Five integrated modes, from LLM annotation to BERT training to inter-annotator validation, with zero programming required.
Articles from 26 Canadian newspapers (1977 to 2024), with 9.2M sentences annotated across 67 hierarchical categories including 8 thematic frames, tone, messengers, events, urgency, and Canadian focus.
A novel method to identify media information cascades through 5 orthogonalized z-score signals, PELT changepoint detection, and a multidimensional score with 17 sub-indices.
A real cascade explored in depth on the Health frame (summer 2023). 87-day cascade, 175 articles, 5,253 annotated sentences, with interactive cluster exploration.
Interactive Presentation
Projects and Resources
All tools, data, and platforms presented in this talk are open source and freely available.