Research Publications
Advancing Natural Language Processing, Computer Vision, and Machine Learning
2025
BnMMLU: Measuring Massive Multitask Language Understanding in Bengali
Saman Sarker Joy
arXiv preprint
This paper introduces BnMMLU, a massive multitask language understanding benchmark designed for Bengali, addressing the gap in low-resource language evaluation. The benchmark spans 23 domains and is used to evaluate several large language models, revealing significant performance differences and the need for improved Bengali-specific training. We release the dataset and benchmark results to encourage further research.
Medical Report Generation and Diagnosis using Multimodal Data and Large Language Models
Saman Sarker Joy
In Progress
This research focuses on generating medical reports by synthesizing multimodal data, including patient history, physician notes, and imaging data (e.g., X-rays, MRI scans). Using large language models with integrated image processing, the system aims to enhance diagnostic accuracy and offer data-driven treatment suggestions. The goal is to create an AI-powered assistant capable of aiding medical professionals by automatically producing detailed, contextually accurate reports, which may assist in early diagnosis and provide recommendations based on recognized patterns in historical data.
2024
Gazetteer-Enhanced Bangla Named Entity Recognition with BanglaBERT Semantic Embeddings K-Means-Infused CRF Model
N. Farhan*, Saman Sarker Joy*, T. B. Mannan and F. Sadeque
arXiv preprint
This paper presents a novel approach to Bangla Named Entity Recognition (NER) by combining Gazetteer information with BanglaBERT semantic embeddings and a K-Means-infused CRF model. Our method achieves state-of-the-art performance on standard Bangla NER datasets.
2023
BanglaClickBERT: Bangla Clickbait Detection from News Headlines using Domain Adaptive BanglaBERT and MLP Techniques
Saman Sarker Joy, T. D. Aishi, N. T. Nodi and A. A. Rasel
Proceedings of the 21st Annual Workshop of the Australasian Language Technology Association
We introduce BanglaClickBERT, a novel approach for detecting clickbait in Bangla news headlines using domain-adaptive BanglaBERT and MLP techniques. Our model significantly outperforms existing methods in Bangla clickbait detection.
Feature-Level Ensemble Learning for Robust Synthetic Text Detection with DeBERTaV3 and XLM-RoBERTa
Saman Sarker Joy, T. D. Aishi
Proceedings of the 21st Annual Workshop of the Australasian Language Technology Association
This paper presents a feature-level ensemble approach for synthetic text detection, combining DeBERTaV3 and XLM-RoBERTa models. Our method demonstrates improved robustness in detecting AI-generated text across multiple languages.