projects

A collection of projects I've worked on.

NegativeResultDetector: SciBERT Text Classification Model

SciBERT Model Architecture

NegativeResultDetector is a text classification model based on SciBERT, developed to identify whether scientific abstracts in clinical psychology and psychotherapy report positive results exclusively or contain mixed/negative findings.

This model is featured in the publication “Classifying Positive Results in Clinical Psychology Using Natural Language Processing” by Louis Schiekiera, Jonathan Diederichs, and Helen Niemeyer, published in the ZfP special issue on Natural Language Processing in Psychology.

Data and Methods

  • Annotated dataset of over 1,900 clinical psychology abstracts categorized into:
    • Positive results only
    • Mixed or negative results
  • Model validation performed on clinical psychology data and two psychotherapy datasets.

Performance Metrics

  • Accuracy: 86.4% on the test set
  • F1 Score:
    • Mixed/negative results: 0.867
    • Positive results only: 0.830

Model Usage on Hugging Face

Use the hosted inference API on Hugging Face by entering an abstract or testing provided examples.

📂 View on GitHub

ShootingBias: MATLAB Psychtoolbox Implementation

Shooter Task Example

ShootingBias is a MATLAB Psychtoolbox implementation of the “shooter task,” a well-established behavioral test used in social psychology to measure implicit racial bias (Correll et al., 2002; Essien et al., 2017).

Background

The shooter bias refers to implicit biases revealed by participants’ reaction times and decisions to “shoot” or “not shoot” based on the perceived threat from individuals of different ethnic backgrounds. It assesses the influence of racial stereotypes on rapid decision-making.

Experimental Design

This implementation replicates Essien et al. (2017) using a 2 × 2 repeated-measures design with two factors:

  • Target Ethnicity: Arab-Muslim vs. White
  • Object Type: Armed vs. Unarmed

Hypotheses

  • Reaction times will be faster for armed Arab-Muslim targets compared to armed White targets.
  • Reaction times will be slower for unarmed Arab-Muslim targets compared to unarmed White targets.
  • Participants will demonstrate greater liberal response bias toward Arab-Muslim targets.

References

  • Correll, J., Park, B., Judd, C. M., & Wittenbrink, B. (2002). Journal of Personality and Social Psychology, 83(6), 1314–1329.
  • Essien, I., Stelter, M., Kalbe, F., Koehler, A., Mangels, J., & Meliß, S. (2017). Journal of Experimental Social Psychology, 70, 41–47.

📂 View on GitHub