Longformer-es-m-base-ICE-MR24-DD

Model Card

Model Description

Longformer-es-m-base-ICE-MR24-DD is a Spanish long-context language model for mental disorder detection, trained using the Incremental Context Expansion (ICE) methodology. The model was developed for the MentalRisk 2024 Disorder Detection (MR24-DD) shared task and builds upon the Longformer-es-mental-base foundation model.

This model corresponds to the base-sized version of the ICE-adapted Longformer models. Compared to its large counterpart, it contains fewer parameters, providing a more computationally efficient alternative while preserving the ability to process long user message histories.

The ICE methodology restructures the training data at the context level, enabling the model to learn from progressively expanding user message histories rather than complete user timelines. This training strategy allows the model to better capture early and partial signals of mental health disorders, while remaining effective in full-context (user-level) evaluation settings.

The model is based on the Longformer architecture and supports input sequences of up to 4096 tokens, making it suitable for disorder detection scenarios where evidence is distributed across long and heterogeneous user histories.

• Developed by: ELiRF group, VRAIN (Valencian Research Institute for Artificial Intelligence), Universitat Politècnica de València
• Shared by: ELiRF
• Model type: Transformer-based sequence classification model (Longformer)
• Language: Spanish
• Base model: Longformer-es-mental-base
• License: Same as base model

Uses

This model is intended for research purposes in mental health disorder detection.

Direct Use

The model can be used directly for mental disorder detection from Spanish user-generated content, producing predictions from incrementally available user message histories.

Downstream Use

• Mental disorder detection (user-level classification)
• Early and incremental mental health screening
• Comparative studies on disorder detection methodologies
• Research on temporally-aware and incremental NLP models

Out-of-Scope Use

• Automated intervention systems without human supervision
• Use on languages other than Spanish
• High-stakes decision-making affecting individuals’ health or safety

ICE Methodology

Incremental Context Expansion (ICE) is a training methodology designed for early and incremental mental health detection tasks. Instead of training models on full user histories, ICE generates multiple incremental contexts per user, each corresponding to a partial message history.

This approach enables the model to:

• Learn from incomplete and early evidence
• Improve robustness under early and incremental evaluation metrics
• Better align training conditions with real-world deployment scenarios

ICE modifies the dataset construction process while keeping the standard fine-tuning pipeline unchanged.

Bias, Risks, and Limitations

• Training data originates from social media platforms and may contain demographic, cultural, and linguistic biases.
• Automatically translated texts may introduce translation artifacts or subtle semantic shifts.
• Disorder detection is inherently subjective and context-dependent.
• The model does not provide explanations or clinical interpretations of its predictions.

How to Get Started with the Model

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("ELiRF/Longformer-es-m-base-ICE-MR24-DD")
model = AutoModelForSequenceClassification.from_pretrained(
    "ELiRF/Longformer-es-m-base-ICE-MR24-DD"
)

inputs = tokenizer(
    "Ejemplo de historial de mensajes relacionado con salud mental.",
    return_tensors="pt",
    truncation=True,
    max_length=4096
)

outputs = model(**inputs)

Training Details

Training Data

The model was fine-tuned on the MentalRisk 2024 Disorder Detection (MR24-DD) dataset. Training data was restructured using the ICE methodology, generating incremental user contexts from original user timelines.

Training Procedure

• Base model: Longformer-es-mental-base
• Fine-tuning strategy: ICE-based context-level training
• Objective: Sequence classification
• Training regime: fp16 mixed precision

Evaluation

Results

When evaluated on the MentalRisk 2024 Disorder Detection task, Longformer-es-m-base-ICE-MR24-DD shows competitive performance under both full-context (user-level) and early / incremental detection evaluation settings, offering a favorable trade-off between performance and computational efficiency.

Environmental Impact

• Hardware type: NVIDIA A40 GPUs
• Training time: several hours (fine-tuning)

Technical Specifications

Model Architecture and Objective

• Architecture: Longformer (base)
• Objective: Sequence classification
• Maximum sequence length: 4096 tokens
• Model size: approximately 150M parameters

Citation

This model is part of an ongoing research project. The associated paper is currently under review and will be added to this model card once the publication process is completed.

Model Card Authors

ELiRF research group (VRAIN, Universitat Politècnica de València)