AvinabhDutta-Dev/assamese-sentiment-muril

Assamese Sentiment Analysis — Fine-Tuned MuRIL

This repository contains a fine-tuned MuRIL (Multilingual Representations for Indian Languages) transformer model optimized for binary sentiment analysis on Assamese textual reviews. It maps localized Assamese content into binary sentiment classes (Positive and Negative).

---

Model Summary

• Model Architecture: MuRIL (google/muril-base-cased)
• Task: Binary Sentiment Classification
• Language: Assamese (as)
• Developer: Avinabh Dutta
• Primary Framework: PyTorch & Hugging Face Transformers
• Top Performance: 96.33% Accuracy | 0.9605 Macro F1

---

Performance & Evaluation Metrics

The model was evaluated against a held-out test dataset at step 1800 (Epoch 4.0), delivering strong stability across macro-averaged evaluation dimensions:

Metric	Score	Percentage
Accuracy	0.963293	96.33%
Macro F1-Score	0.960526	96.05%
Macro Precision	0.960819	96.08%
Macro Recall	0.960236	96.02%
ROC-AUC	0.988171	98.82%

---

Dataset Overview

The dataset consists of localized social media commentaries, media assessments, and audience reviews covering drama (নাটক) and cultural content in Assamese.

• Feature Representation: Focuses on extracting linguistic subtext from Assamese textual tokens, tracking features such as character length, word density, punctuation patterns, and emoji occurrences.
• Target Classes: Binary Classification (0: Negative, 1: Positive).

---

How to Use

You can easily load and run inference with this model using the Hugging Face transformers library:

from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline

# Load model and tokenizer
model_name = "AvinabhDutta-Dev/assamese-sentiment-muril"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# Create pipeline
classifier = pipeline("sentiment-analysis", model=model, tokenizer=tokenizer)

# Example text (Assamese)
sample_text = "চলচ্চিত্ৰখনৰ কাহিনী আৰু অভিনয় অতি সুন্দৰ হৈছে।"
result = classifier(sample_text)

print(result)

---

Training Hyperparameters

Fine-tuning parameters were tightly managed across a strict execution run:

Parameter	Value
Total Epochs	4.0
Train Batch Size	16
Max Training Steps	1800
Peak Learning Rate	2e-05 (decayed to 1.234e-08)
Logging Steps	100
Evaluation Steps	500

---

Limitations & Edge Cases

• Syntactic Shifts & Contrast Markers: The model can face challenges when encountering nested sarcasm, hard negation shifts, or complex contrast structures.
• Contrast Words: Sentences containing explicit positive tokens paired with contrast markers (e.g., কিন্তু [but], কাৰণ [because]) may occasionally trigger false positives. Manual rule overrides or contrastive data augmentation are recommended for edge-case resolution.

---

Citation & Contact

If you use this model or refer to this work in your research, please link back to this Hugging Face repository or the GitHub project repository. If you use this model, please cite the associated preprint:

Dev, C., Baishya, B., Dey, N., Dutta, A., & Dutta, S. (2026). Sentiment Analysis of Assamese Reviews using Transfer Learning Models. Preprint. https://doi.org/10.2139/ssrn.6357982