Unifying the Extremes: Developing a Unified Model for Detecting and Predicting Extremist Traits and Radicalization

Setup Instructions

This project uses a Conda environment for managing dependencies. To create and activate the environment:

# Clone the repository
git clone https://github.com/humanlab/extremism.git
cd extremism

# Create the Conda environment
conda env create -f environment.yml

# Activate the environment
conda activate extremism

Please install PyTorch

Item Scoring

Scale	Source	Construct(s) Assessed	# of Items	bibkey
Extremism Scale	Ozer and Bertelsen (2018)	General Extremist Attitudes	14	ozer-2018-capturing
Social Dominance Orientation	Ho et al. (2015)	Preferences for Social Hierarchy	8	ho2015nature
Radicalism Intention	Moskalenko & McCauley (2009)	Propensity for Radical Action	4	moskalenko2009measuring
Violent Intention	Obaidi et al. (2018a, 2018b)	Likelihood of Violent Behavior	7	obaidi2018living, obaidi2018mistreatment
Nationalism Scale	Weiss (2003)	National Identity and Superiority	4	weiss2003cross
Right-Wing Authoritarianism	Zakrisson (2005)	Obedience to Authority and Tradition	15	zakrisson2005construction
Self-Categorization Scale	Ellemers et al. (1999)	Group Identity and Affiliation	3	ellemers1999self
Dirty Dozen	Jonason & Webster (2010)	"Dark" Personality Traits	12	jonason2010dirty
General Extremist	Jungkunz et al. (2024)	Broad Extremist Tendencies	5	jungkunz2024measuring
Left-Wing Radical	Jungkunz et al. (2024)	Radical Left Ideological Views	6	jungkunz2024measuring
Right-Wing Radical	Jungkunz et al. (2024)	Radical Right Ideological Views	7	jungkunz2024measuring
Ethnic Intolerance	Weiss (2003)	Xenophobic Attitudes	4	weiss2003cross

Data

Option 1: Use Our Preprocessed Datasets

Request access to the datasets: [data request link]
Place the CSV files in: data/final_dataset/
Use the --data flag with one of the dataset names below:

python item_scoring.py --data sample

Dataset Name	Description
`sample`	5 random posts from Reddit (for demo)
`stormfront`	White supremacist forum posts
`poli`	Political discussion posts
`incel`	Involuntary celibate forum posts
`control`	Baseline/control forum posts
`isis`	Jihadist forum posts
`incel_history`	User histories before joining incel
`random`	Large set of random online posts

Option 2: Use Your Own Dataset

To use your own dataset, just pass the file path and the name of the text column using --custom-dataset and --text-column:

python item_scoring.py \
  --custom-dataset path/to/your_file.csv \
  --text-column name_of_text_column

Choosing a Scale

You can choose one or more scales used in our paper (stored in data/scales/) by name with the --scale flag. Multiple scales can be passed to run together in one job. You can also use your own custom scale by passing in a .txt file path. Each scale file should list the scale items, one per line.

python item_scoring.py --scale jungkunz_gex --scale RWA --scale /path/to/my_scale.txt

To replicate the Extremism Eleven Factors, use all the scales using all as the flag: --scale all. This will extract all the extremism scale similarity scores and save to a single file.

python item_scoring.py --scale all

Saving Output

Output will be saved automatically in data/irt_data/ unless you specify --output.

To save somewhere else, use the --output option to specify the desired file path.
If the file already exists and you want to overwrite it, the script requires you to use -w.

Extract Factor Scores

This module provides functionality for applying factor analysis models to extract factor scores from item-scored similarity data.

The extract_factor_scores.py script is designed to:

Load a pre-trained 11-factor extremism model
Apply the factor model to item-scored similarity data
Generate factor scores for psychological constructs related to extremism
Save the resulting factor scores for downstream analysis

Applying the Factor Model:

Fucntion apply_factor_model(fa, df):

Purpose: Loads a pre-trained factor analysis model from a pickle file and applies it to transform input data into factor scores
Parameters:
- factor_model_path: ath to the model file. Defaults to "model/extremist_11_factor_model.pkl"
- df: pandas DataFrame containing the similarity scores to transform
Returns: numpy array of factor scores

Training a new model:

Function train_and_save_factor_model(df, n_factors, negative_code_cols, model_path):

Purpose: Trains a new factor analysis model and saves it to disk
Parameters:
- df: pandas DataFrame with training data
- n_factors (int, optional): Number of factors to extract. Defaults to 11
- negative_code_cols (list, optional): Columns that need reverse coding
- model_path (str, optional): Where to save the new model. Defaults to "model/new_factor_model.pkl"
Returns: Fitted FactorAnalyzer object
Note: Applies reverse coding to specified columns before training

The Pipeline

To get the scores for downstream analysis:

Item Scoring: Run item_scoring.py to generate similarity scores
Factor Extraction: Run extract_factor_scores.py to convert similarities to factor scores
Analysis: Use the factor scores for downstream psychological analysis

Citation

If you use our code, please cite our work. Note: Allison Lahnala and Vasudha Varadarajan contributed equally to this work (co-first authors).

@article{Lahnala_Varadarajan_Flek_Schwartz_Boyd_2025,
  title={Unifying the Extremes: Developing a Unified Model for Detecting and Predicting Extremist Traits and Radicalization},
  volume={19},
  url={https://ojs.aaai.org/index.php/ICWSM/article/view/35860},
  DOI={10.1609/icwsm.v19i1.35860},
  note={Allison Lahnala and Vasudha Varadarajan contributed equally to this work.},
  number={1},
  journal={Proceedings of the International AAAI Conference on Web and Social Media},
  author={Lahnala, Allison and Varadarajan, Vasudha and Flek, Lucie and Schwartz, H. Andrew and Boyd, Ryan L},
  year={2025},
  month={Jun.},
  pages={1051--1067}
}

Please also cite the papers that introduced the scales used in this work.

@article{ozer-2018-capturing,
    author = {Ozer, Simon and Bertelsen, Preben},
    title = {Capturing violent radicalization: Developing and validating scales measuring central aspects of radicalization},
    journal = {Scandinavian Journal of Psychology},
    volume = {59},
    number = {6},
    pages = {653-660},
    year = {2018}
}
@article{ho2015nature,
    title={The nature of social dominance orientation: Theorizing and measuring preferences for intergroup inequality using the new SDO₇ scale.},
    author={Ho, Arnold K and Sidanius, Jim and Kteily, Nour and Sheehy-Skeffington, Jennifer and Pratto, Felicia and Henkel, Kristin E and Foels, Rob and Stewart, Andrew L},
    journal={Journal of personality and social psychology},
    volume={109},
    number={6},
    pages={1003},
    year={2015},
    publisher={American Psychological Association}
}
@article{moskalenko2009measuring,
    title={Measuring political mobilization: The distinction between activism and radicalism},
    author={Moskalenko, Sophia and McCauley, Clark},
    journal={Terrorism and political violence},
    volume={21},
    number={2},
    pages={239--260},
    year={2009},
    publisher={Taylor \& Francis}
}
@article{obaidi2018living,
    title={Living under threat: Mutual threat perception drives anti-Muslim and anti-Western hostility in the age of terrorism},
    author={Obaidi, Milan and Kunst, Jonas R and Kteily, Nour and Thomsen, Lotte and Sidanius, James},
    journal={European Journal of Social Psychology},
    volume={48},
    number={5},
    pages={567--584},
    year={2018},
    publisher={Wiley Online Library}
}
@article{obaidi2018living,
    title={Living under threat: Mutual threat perception drives anti-Muslim and anti-Western hostility in the age of terrorism},
    author={Obaidi, Milan and Kunst, Jonas R and Kteily, Nour and Thomsen, Lotte and Sidanius, James},
    journal={European Journal of Social Psychology},
    volume={48},
    number={5},
    pages={567--584},
    year={2018},
    publisher={Wiley Online Library}
}
@article{weiss2003cross,
    title={A cross-national comparison of nationalism in Austria, the Czech and Slovac Republics, Hungary, and Poland},
    author={Weiss, Hilde},
    journal={Political Psychology},
    volume={24},
    number={2},
    pages={377--401},
    year={2003},
    publisher={Wiley Online Library}
}
@article{zakrisson2005construction,
    title={Construction of a short version of the Right-Wing Authoritarianism (RWA) scale},
    author={Zakrisson, Ingrid},
    journal={Personality and individual differences},
    volume={39},
    number={5},
    pages={863--872},
    year={2005},
    publisher={Elsevier}
}
@article{ellemers1999self,
    title={Self-categorisation, commitment to the group and group self-esteem as related but distinct aspects of social identity},
    author={Ellemers, Naomi and Kortekaas, Paulien and Ouwerkerk, Jaap W},
    journal={European journal of social psychology},
    volume={29},
    number={2-3},
    pages={371--389},
    year={1999},
    publisher={Wiley Online Library}
}
@article{jonason2010dirty,
    title={The dirty dozen: a concise measure of the dark triad.},
    author={Jonason, Peter K and Webster, Gregory D},
    journal={Psychological assessment},
    volume={22},
    number={2},
    pages={420},
    year={2010},
    publisher={American Psychological Association}
}
@article{jungkunz2024measuring,
    title={Measuring political radicalism and extremism in surveys: Three new scales},
    author={Jungkunz, Sebastian and Helbling, Marc and Osenbr{\"u}gge, Nina},
    journal={Plos one},
    volume={19},
    number={5},
    pages={e0300661},
    year={2024}
}

Name		Name	Last commit message	Last commit date
Latest commit History 14 Commits
data		data
item-scoring		item-scoring
model		model
.gitignore		.gitignore
README.md		README.md
environment.yml		environment.yml

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Repository files navigation

Unifying the Extremes: Developing a Unified Model for Detecting and Predicting Extremist Traits and Radicalization

Setup Instructions

Item Scoring

Data

Option 1: Use Our Preprocessed Datasets

Option 2: Use Your Own Dataset

Choosing a Scale

Saving Output

Extract Factor Scores

Applying the Factor Model:

Training a new model:

The Pipeline

Citation

About

Uh oh!

Releases

Packages

Languages

humanlab/extremism

Folders and files

Latest commit

History

Repository files navigation

Unifying the Extremes: Developing a Unified Model for Detecting and Predicting Extremist Traits and Radicalization

Setup Instructions

Item Scoring

Data

Option 1: Use Our Preprocessed Datasets

Option 2: Use Your Own Dataset

Choosing a Scale

Saving Output

Extract Factor Scores

Applying the Factor Model:

Training a new model:

The Pipeline

Citation

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Languages

Packages