ordered-HC-gemma3-4b-fine-tuned

Model description

This model is one of two population-specific models created to study linguistic differences between healthy controls and patients with psychosis. The general idea of the project is to compare both models using perplexity and other language-based signals. This is the Healthy model for simulating responses from healthy control participants in clinical interviews.

Training data

The model was trained on question-answer pairs extracted from the Discourse-UWO clinical interview dataset.

Intended use

This model is intended for research in clinical NLP, language modeling, and psychosis-related speech analysis. It can also be used for experiments with population-specific text generation.

Limitations

This model is for research purposes only and is not intended for diagnosis or clinical decision-making. The dataset is limited and imbalanced, so outputs may reflect dataset-specific biases.

Citation

If you use this model, cite the associated thesis documentation:

@misc{cano2026population_modeling_psychosis,
  title={Towards Explainable AI for Psychosis Detection Through Clinical Language Modeling},
  author={Pablo Ramon Cano Ortiz},
  year={2026},
  note={Undergraduate thesis, Universidad Politecnica de Madrid}
}