This repository contains a fine-tuned IndicBERT (ai4bharat/indic-bert) 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: IndicBERT (ai4bharat/indic-bert)
Task: Binary Sentiment Classification
Language: Assamese (as)
Developer: Avinabh Dutta
Primary Framework: PyTorch & Hugging Face Transformers
Top Performance: 92.71% Accuracy | 0.9220 Macro F1

Performance & Evaluation Metrics

The model was evaluated against a held-out test dataset, demonstrating stable generalization across all macro-averaged dimensions:

Metric	Score	Percentage / Value
Accuracy	`0.927141`	92.71%
Macro F1-Score	`0.921967`	0.9220
Macro Precision	`0.920584`	0.9206
Macro Recall	`0.923433`	0.9234
ROC-AUC	`0.977318`	0.9773
Cohen's Kappa	`0.843940`	0.8439

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: Captures 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:

python

from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline

# Load model and tokenizer
model_name = "AvinabhDutta-Dev/assamese-sentiment-indicbert"
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 managed alongside the overall transformer execution pipeline:

Parameter	Value
Base Model	`ai4bharat/indic-bert`
Total Epochs	`4.0`
Train Batch Size	`16`
Learning Rate	`2e-05`
Weight Decay	`0.01`
Warmup Ratio	`0.1`
Checkpoint Selection	Best Macro F1

Limitations & Edge Cases

Structural Capacity: Due to its compact architecture and generalized multi-language Indic representation layer, IndicBERT captures fewer domain-specific nuances compared to larger models like MuRIL.
Contextual Shifts & Boundary Markers: The model shows higher vulnerability when processing mixed token properties, hard sentence boundaries, or implicit sentiment reversals. Sentences weaving complex local structural markers or contrast transitions may result in false predictions.

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