README.md

---
license: apache-2.0
datasets:
- pratikkalamkar/Movie_Scripts_with_Age_Ratings_by_Pratik_Kalamkar-Hi
language:
- hi
metrics:
- accuracy
- f1
- precision
- recall
base_model:
- distilbert/distilbert-base-multilingual-cased
pipeline_tag: text-classification
library_name: transformers
tags:
- movie-certification
- content-rating
- movie-script
- nlp
- text-classification
- multilingual
- distilbert
- knowledge-distillation
- long-document
- explainable-ai
---


# MovieCert Hindi Teacher Model

## Model Details

### Model Description

This repository contains the Hindi Teacher model developed for the research paper **"Lightweight and Explainable Neural Models for Multilingual Movie Script Certification."**

The model is fine-tuned from the multilingual DistilBERT architecture to classify Hindi movie scripts into age-rating categories. It serves as the teacher model in a knowledge distillation framework used to develop lightweight student models for efficient deployment.

- **Developed by:** Pratik N. Kalamkar
- **Funded by:** Self-funded academic research
- **Shared by:** Pratik N. Kalamkar
- **Model type:** Transformer-based sequence classification model (DistilBERT)
- **Language(s):** Hindi
- **License:** Apache 2.0
- **Finetuned from model:** distilbert-base-multilingual-cased

---

## Model Sources

- **Repository:** https://github.com/pratik1986/Lightweight-and-Explainable-Neural-Models-for-Multilingual-Movie-Script-Certification
- **Paper:** https://doi.org/10.5815/ijitcs.2026.02.09
- **Dataset:** https://huggingface.co/datasets/pratikkalamkar/Movie_Scripts_with_Age_Ratings_by_Pratik_Kalamkar-Hi

---

# Uses

## Direct Use

This model is intended for automated classification of Hindi movie scripts into age-rating categories.

Potential applications include:

- Automated movie certification research
- Movie script analysis
- Content moderation
- Academic NLP research
- Benchmarking multilingual transformer models

---

## Downstream Use

The model may be used as:

- Teacher model for knowledge distillation
- Base model for further fine-tuning
- Feature extractor
- Long-document text classification research

---

## Out-of-Scope Use

This model is **not** intended for:

- Legal or official movie certification decisions
- Safety-critical systems
- Hate speech detection
- Sentiment analysis
- Medical or legal decision making

Predictions should not replace human review.

---

# Bias, Risks and Limitations

The model was trained using a curated dataset of Hindi movie scripts and inherits limitations of the dataset, including:

- Limited dataset size
- Possible class imbalance
- Genre imbalance
- Cultural bias
- Language-specific ambiguity

Performance may decrease on scripts that differ substantially from the training distribution.

---

## Recommendations

The model should be used as an assistive tool rather than a replacement for human reviewers.

---

# Training Details

## Training Data

The model was trained using the **Hindi Movie Scripts with Verified Age Ratings** https://huggingface.co/datasets/pratikkalamkar/Movie_Scripts_with_Age_Ratings_by_Pratik_Kalamkar-Hi dataset curated by the author.

Training data consists of Hindi movie scripts labelled with their corresponding age ratings.

---

## Training Procedure

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

### Preprocessing

- UTF-8 normalization
- Tokenization using DistilBERT tokenizer
- Sliding-window chunking for long movie scripts
- Chunk-level prediction aggregation

---

### Training Hyperparameters

- Base model: distilbert-base-multilingual-cased
- Epochs: 6
- Batch size (train): 8
- Batch size (evaluation): 16
- Learning rate: 2e-5
- Weight decay: 0.01
- Warmup steps: 50
- Optimizer: AdamW
- Maximum gradient norm: 1.0
- Early stopping patience: 2
- Mixed precision: Disabled (FP32)

---

# Evaluation

## Testing Data

Held-out Hindi movie script test split.

---

## Metrics

The following metrics were used:

- Accuracy
- Precision
- Recall
- Macro F1-score
- Confusion Matrix

---

## Results

The complete experimental results are reported in the associated journal paper.

---

# Technical Specifications

## Model Architecture

- DistilBERT multilingual encoder
- Sequence Classification Head
- Fine-tuned on Hindi movie scripts

---

## Software

- Python
- PyTorch
- Hugging Face Transformers
- Scikit-learn

---

# Citation

If you use this model in your research, please cite:

**BibTeX**

```bibtex
@article{kalamkar2026moviecertification,
  author = {Pratik N. Kalamkar and Prasadu Peddi and Yogesh K. Sharma},
  title = {Lightweight and Explainable Neural Models for Multilingual Movie Script Certification},
  journal = {International Journal of Information Technology and Computer Science},
  volume = {18},
  number = {2},
  pages = {146--160},
  year = {2026},
  doi = {10.5815/ijitcs.2026.02.09}
}
```

---

# More Information

This model is one component of the MovieCert multilingual movie certification framework comprising:

- English Teacher Model
- Hindi Teacher Model
- Marathi Teacher Model


---

# Model Card Author

Pratik N. Kalamkar

---

# Contact

For questions regarding the model or the associated research, please contact:

Pratik N. Kalamkar

GitHub: https://github.com/pratik1986