dataset_info:
features:
- name: source
dtype: string
- name: text
dtype: string
splits:
- name: train
num_bytes: 1636020
num_examples: 1982
download_size: 944079
dataset_size: 1636020
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
task_categories:
- text-classification
language:
- en
tags:
- medical
PSYOP Detection Dataset
Dataset Description
The PSYOP Detection Dataset is a collection of textual excerpts extracted from academic and military literature discussing psychological operations (PSYOP), information warfare, propaganda, and psychological influence strategies.
The dataset was created to support research on detecting psychological manipulation techniques in text, including persuasion, propaganda, and influence operations.
Source documents were segmented into smaller text chunks to facilitate training and fine-tuning of language models for classification, detection, or analysis tasks related to psychological operations.
Dataset Structure
The dataset contains the following columns:
- source: The original PDF document the text was extracted from.
- text: A chunk of text extracted from the source document.
Each row represents a text segment approximately 800 characters long.
Total rows: ~1982
Data Collection Process
The dataset was constructed by:
- Identifying publicly available academic and research documents discussing psychological operations, propaganda, and information warfare.
- Discovering and sourcing relevant literature with the assistance of research tools including Google Gemini and Consensus.app, which were used to identify relevant academic publications and research material.
- Extracting textual content from the selected PDF documents.
- Segmenting the extracted text into smaller chunks suitable for machine learning training and analysis.
The goal of this preprocessing was to produce training examples that models can learn from when identifying patterns associated with psychological influence operations.
Data Provenance and Responsibility Disclaimer
The documents used to construct this dataset were identified as publicly accessible academic or research materials at the time of collection. The dataset contains only extracted textual segments intended for research, machine learning experimentation, and academic analysis.
The dataset creator does not claim ownership of the original source materials. All intellectual property rights remain with the respective authors, publishers, or rights holders.
If any rights holder believes that material included in this dataset should not be redistributed or used in this context, they are encouraged to contact the dataset maintainer through Hugging Face so that the relevant content can be reviewed and removed if appropriate.
The dataset is provided as-is for research and educational purposes, and the creator assumes no liability for unintended use or misinterpretation of the included material.
Intended Uses
This dataset may be used for:
- Fine-tuning language models to detect persuasion or propaganda techniques
- Research on psychological operations and information warfare
- NLP research on manipulation detection
- Training classifiers for PSYOP-related content analysis
Limitations
This dataset contains only excerpts from documents describing psychological operations rather than labeled examples of PSYOP content itself.
As a result:
- The dataset may not fully represent real-world PSYOP messaging.
- Additional labeling may be required for supervised learning tasks.
- The dataset is primarily intended as a corpus for further annotation or feature extraction.
Ethical Considerations
The dataset contains discussions of psychological manipulation and influence operations. It should be used responsibly and primarily for research aimed at detecting or mitigating harmful information campaigns.
Citation
If you use this dataset in research, please cite:
Giroux, T. (2026). PSYOP Detection Dataset. Hugging Face.