CCF Database: A Machine-Learning-Annotated Corpus of 266,271 Canadian Climate Articles (1978–2024) — PostgreSQL edition
Authors/Creators
- 1. Université de Sherbrooke / Université de Montréal
- 2. Université de Montréal
Description
The Canadian Climate Framing (CCF) Database is a comprehensive, machine-learning-annotated corpus of climate-change media coverage in Canada. It comprises 266,271 articles from 20 major Canadian newspapers (1978-2024) processed into 9,198,958 two-sentence analytical units (82.9% English, 17.1% French). Each unit is annotated across 65 hierarchical categories by 128 BERT and CamemBERT classifiers, with a macro F1 of 0.866 on a 1,000-sentence gold standard double-coded by an independent annotator (Gwet's AC1 = 0.894, Krippendorff's α = 0.698, Cohen's κ = 0.596 on the 400 blind sentences). Each category receives an A/B/C reliability tier summarising annotation quality from classifier performance and inter-coder agreement. The deposit ships six relational tables (bibliographic metadata, sentence-level annotations, named-entity rollups, article-level aggregates, per-category reliability tiers, and 9,462,845 BAAI/bge-m3 sentence-and-title embeddings). Raw newspaper text is excluded for copyright reasons; bibliographic coordinates (media, date, title, author, page_number) are sufficient for any researcher with institutional access to Factiva, Eureka.cc or ProQuest Canadian Major Dailies to recover the original sentences. This deposit accompanies a methodology paper currently under revision at Scientific Data (Nature Portfolio).
This deposit is the canonical PostgreSQL edition. It contains a pg_dump -Fd directory archive (compressed into a single .tar file) of the six relational tables, including the pgvector extension and HNSW cosine indexes for sub-second semantic-similarity search. Restoration is a one-liner:
tar -xf CCF_Database.tar && createdb CCF_Database && psql -d CCF_Database -c 'CREATE EXTENSION IF NOT EXISTS vector;' && pg_restore -d CCF_Database --no-owner --no-privileges -j 8 CCF_Database_dump
A column-oriented Apache Parquet mirror of the same six tables is available as the sister deposit on Zenodo (cross-referenced in Related identifiers). The Parquet mirror is recommended for users without PostgreSQL access (it is directly readable by pandas, polars, R/arrow, DuckDB, and Spark).
The full annotation pipeline, training data, manual-annotation JSONL, intercoder-reliability benchmark, methodology manuscript (LaTeX sources + PDF), and reproducibility scripts are bundled with this deposit as ccf_code_and_paper.tar.gz. The same materials are also available on the project's OSF companion deposit (10.17605/OSF.IO/Q5W47) and on the development mirror at GitHub.
Requirements: PostgreSQL 16 or 17 with pgvector ≥ 0.8.2 (for halfvec(1024) storage of the sentence embeddings).
Notes
Version 1.1.0 (2026-06-12): schema-level update to v1.0.0.
- Column rename in
CCF_article_aggregates:dominant_frame→top_frame;dominant_frame_prop→top_frame_prop. The semantics are unchanged (the frame with the highestprop_Xin the article); the rename avoids a collision with the meaning that 'dominant frame' carries in a separate paper by the same authors. - Throughout the manuscript and documentation, 'thematic frames' is replaced by 'main frames' to remove the collision with the narrower definition of 'thematic frame' in part of the framing literature. The eight main frames are unchanged.
- Reproducible tables (Supplementary Tables S3-S12, Data Overview tables) and the CODEBOOK regenerated from the renamed schema; Supplementary Table S3 'Thematic frames' section header renamed to 'Main frames'.
- Manuscript abstract revised for clarity; new Acknowledgements section added.
- Row counts, statistical results, and trained models are byte-identical to v1.0.0.
The original v1.0.0 DOIs (10.5281/zenodo.20346364 and 10.5281/zenodo.20346373) remain valid for v1.0.0 of the deposit (with the original dominant_frame column).
Methods
Climate-related newspaper articles published between 1978 and 2024 in 20 major Canadian outlets (national, regional, and French-language) were retrieved from Factiva, Eureka.cc, and ProQuest Canadian Major Dailies through the institutional subscriptions of Université de Montréal and Université de Sherbrooke. After language detection, deduplication, and a 100-word minimum-length filter, the 266,271 articles were segmented into 9,198,958 two-sentence analytical units with spaCy.
Each unit was annotated across 65 hierarchical binary categories covering eight main frames (Economic, Health, Security, Justice, Political, Scientific, Environmental, Cultural), actors/messengers, events, solutions, emotional tone, geographic focus and urgency. Annotation relies on 128 transformer-based classifiers (BERT for English, CamemBERT for French) trained on more than 4,000 expert-coded sentences, with a reinforced-training phase triggered for low-F1 categories. Models reach a macro F1 of 0.866 on a stratified 1,000-sentence gold standard double-coded by an independent annotator (Gwet's AC1 = 0.894, Krippendorff's α = 0.698, Cohen's κ = 0.596 on the 400 blind sentences).
Per-sentence named entities (persons, organisations, locations) are extracted with a hybrid spaCy + BERT pipeline. Article-level rollups (top frame, framing entropy, deduplicated named-entity arrays) are materialised on PostgreSQL together with 9,462,845 BAAI/bge-m3 sentence-and-title embeddings stored as halfvec(1024) with HNSW cosine indexing through pgvector. A per-category A/B/C reliability tier, jointly determined by classifier macro F1 and inter-coder agreement, accompanies every annotation.
Files
CCF_Methodology.pdf
Files
(39.8 GB)
Additional details
Related works
- Is source of
- 10.5281/zenodo.20667154 (DOI)
- Is supplement to
- 10.17605/OSF.IO/Q5W47 (DOI)
Dates
- Collected
-
1978-01-01/2024-12-31Publication dates of the 266,271 climate-related newspaper articles in the CCF corpus.
References
- Lemor, A., Pillod, A., Taylor, M., & Nadeau, R. (2026). The Canadian Climate Framing (CCF) database: a sentence-level annotated corpus for the analysis of climate-change discourse in the Canadian press. Scientific Data (under revision).