UDDIPOK: A reading comprehension based question answering dataset in Bangla language.

Area of Science:

• Natural Language Processing (NLP)
• Machine Learning
• Deep Learning

Background:

• Reading comprehension (RC) research in Bangla is growing, but lacks original, diverse datasets.
• Existing Bangla RC datasets are often translated, leading to inaccuracies and data anomalies.
• There is a need for a high-quality, native Bangla dataset for NLP tasks.

Purpose of the Study:

• To introduce UDDIPOK, a novel, open-domain Bangla reading comprehension dataset.
• To provide a resource for advancing Bangla NLP research, particularly in machine and deep learning.
• To overcome the limitations of existing translated datasets.

Main Methods:

• The UDDIPOK dataset was created using diverse Bangla sources, including textbooks, school exams, and newspapers.
• It comprises 270 reading passages, 3636 questions, and corresponding answers.
• The dataset is formatted in CSV for easy handling in machine learning research.

Main Results:

• UDDIPOK offers a comprehensive collection of original Bangla reading comprehension data.
• The dataset's diverse origins ensure broad applicability across various NLP tasks.
• CSV formatting facilitates efficient data processing for researchers.

Conclusions:

• UDDIPOK is a valuable, open-domain resource for Bangla Natural Language Processing.
• This dataset will support the development of more accurate and robust Bangla RC models.
• It promotes further research in machine learning and deep learning for the Bangla language.