chrissoria/trump-truth-social

Trump Truth Social Posts Archive

Public posts ("Truths") by Donald J. Trump on Truth Social, enriched with market data, geopolitical event indicators, and LLM-based post classifications. Collected for academic research purposes.

Dataset Description

• —Source: CNN/Stiles Truth Social Archive (live-updating public archive)
• —Posts: ~32,000+ (growing)
• —Date range: February 2022 – present
• —Update frequency: Daily (Truth Social), weekly (all other sources)
• —Maintainer: Chris Soria (UC Berkeley)

Fields

Post metadata

Engagement

Media

Post classification (5-model ensemble)

LLM-classified post categories using a 5-model unanimous-vote ensemble (Llama 4 Maverick, Qwen3-32B, Claude 3 Haiku, GPT-4o-mini, Gemini 2.0 Flash). Multi-label: a post can belong to multiple categories. Available for posts with text since Nov 5, 2024 (election day onwards). Values: 1 = present, 0 = not present.

Market data (18 tickers)

Each ticker has 7 columns following the pattern {ticker}_{metric}. Daily open/close prices are available for all posts. Intraday prices (1hr before through 1hr after) use the highest available resolution: 1-minute (last ~7 days), 5-minute (last ~60 days), or hourly (last ~2 years). Weekend/holiday posts use the most recent trading day. The sp500_resolution column indicates the intraday data resolution.

Metrics per ticker:

Tickers:

GDELT geopolitical events (daily)

Daily aggregates of US-involved events from the GDELT Project via BigQuery. Each row gets the event counts for its post date. Based on CAMEO event coding of global news coverage.

Note: GDELT daily exports are typically available with a ~1 day lag. Posts from the most recent day may have null GDELT columns until the next daily update backfills them.

Raw counts:

Derived:

Available for: military, sanctions, threat, protest, force_posture, diplomatic (pct); military, sanctions, threat, protest, material_conflict (zscore and delta); goldstein_avg, avg_tone (delta).

Intended Use

This dataset is intended for academic research in political science, computational social science, NLP, finance, and related fields. Example use cases:

• —Analyzing the relationship between presidential social media activity and market movements
• —Studying the timing and framing of aggressive policy announcements
• —Discourse analysis and political communication research
• —Event-driven analysis correlating posts with GDELT geopolitical indicators
• —Accessibility research using AI-generated image descriptions

Fair Use Notice

This dataset is compiled from publicly available posts by a public figure for academic research purposes under fair use (17 U.S.C. § 107). The data consists of factual records of public political speech. Source data is from the CNN/Stiles public archive. Market data sourced from Yahoo Finance via yfinance. Geopolitical data from the GDELT Project. Multiple peer-reviewed publications have established precedent for academic use of Truth Social data (see ICWSM 2023, arXiv:2411.01330).

Citation

If you use this dataset in your research, please cite this dataset and the underlying data sources:

This dataset

@misc{soria2026trump_truth_social,
  title={Trump Truth Social Posts Archive},
  author={Soria, Christopher},
  year={2026},
  publisher={HuggingFace},
  url={https://huggingface.co/datasets/chrissoria/trump-truth-social}
}

Source data: Truth Social posts

The raw post data is sourced from Matt Stiles' CNN Truth Social archive:

@misc{stiles2024truthsocial,
  title={Trump Truth Social Archive},
  author={Stiles, Matt},
  year={2024},
  publisher={CNN},
  url={https://github.com/stiles/trump-truth-social-archive}
}

Market data: Yahoo Finance

Stock and ETF price data is sourced from Yahoo Finance via the yfinance Python library:

@software{yfinance,
  title={yfinance: Download market data from Yahoo! Finance API},
  author={Aroussi, Ran},
  url={https://github.com/ranaroussi/yfinance},
  license={Apache-2.0}
}

Geopolitical events: GDELT Project

Daily geopolitical event aggregates are sourced from the GDELT Project:

@article{leetaru2013gdelt,
  title={GDELT: Global Data on Events, Location and Tone, 1979--2012},
  author={Leetaru, Kalev and Schrodt, Philip A.},
  journal={ISA Annual Convention},
  year={2013},
  url={https://www.gdeltproject.org/}
}

LLM classification and image descriptions

Post classifications were generated using cat-stack with a 5-model ensemble (Llama 4 Maverick, Qwen3-32B, Claude 3 Haiku, GPT-4o-mini, Gemini 2.0 Flash). Image descriptions were generated using Qwen2.5-VL-72B.

@software{soria2026catstack,
  title={cat-stack: Domain-agnostic text, image, and PDF classification engine powered by LLMs},
  author={Soria, Christopher},
  year={2026},
  url={https://github.com/chrissoria/cat-stack}
}

Part of the cat-pol ecosystem

This dataset is part of the cat-pol political text analysis toolkit. Install with:

pip install "cat-pol[sources]"

from cat_pol.sources import fetch_trump_truths
df = fetch_trump_truths(since="2024-01-01")