bilalzafar/IslamicBank-HCDi-RoBERTa

IslamicBank-HCDi-RoBERTa

Model description

IslamicBank-HCDi-RoBERTa is a RoBERTa-based NLP transformer model fine-tuned to classify dimensions of human capital disclosure in Islamic banking. The model identifies whether a text segment is substantively related to human capital and assigns it to one of six human capital disclosure dimensions or to a separate Not related class.

The model is based on FacebookAI/roberta-base and was fine-tuned using a supervised seven-class text classification design. It is intended for research on human capital disclosure, Islamic banking, sustainability reporting, corporate disclosure, workforce disclosure and computational text analysis.

Labels and Training data

The model was fine-tuned on a labelled text classification dataset developed from Islamic banking disclosure text. Candidate text segments were extracted using a human capital lexicon and then classified into human capital disclosure dimensions.

The final labelled dataset contained 13,218 observations across seven classes:

The dataset was split using a stratified 80:10:10 train-validation-test design:

Training procedure

The model was fine-tuned using the Hugging Face Transformers library.

Class-weighted cross-entropy loss was used to address class imbalance.

Evaluation results

The selected RoBERTa model achieved the following performance:

Class-wise test performance

Model comparison

Four transformer models were fine-tuned and compared under the same experimental setup.

RoBERTa base was selected because it achieved the highest validation macro-F1 and strongest overall test performance.

GitHub repository and related model

Pipeline notebooks, data preparation files, outputs and reproducible experiments are available on GitHub:

GitHub: bilalezafar/IslamicBank-HCDi-RoBERTa

A related binary human capital classifier is also available on Hugging Face. This model classifies text into two labels: HC and Non-HC.

Related model: bilalzafar/HC-BERT-IFIs

Citation

If you use this model, please cite:

@misc{zafar2026_islamicbank_hcdi_roberta,
  title        = {IslamicBank-HCDi-RoBERTa: A Transformer Model for Classifying Dimensions of Human Capital Disclosure in Islamic Banking},
  author       = {Zafar, Muhammad Bilal},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/bilalzafar/IslamicBank-HCDi-RoBERTa}}
}

How to use

from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline

model_id = "bilalzafar/IslamicBank-HCDi-RoBERTa"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)

classifier = pipeline(
    "text-classification",
    model=model,
    tokenizer=tokenizer,
    top_k=None,
    truncation=True,
    max_length=128
)

text = "The Bank provides regular training programmes to enhance employees' professional and Shariah-related competencies."

predictions = classifier(text)

best_prediction = max(predictions[0], key=lambda x: x["score"])

print(f"Dimension: {best_prediction['label']} | Prediction score: {best_prediction['score']:.4f}")

# Output {Dimension: Training, Development, and Human Capital Investment | Prediction score: 0.9986}