Shongbad Chitro: A Multimodal Bangla News Image-Text Dataset
Description
Shongbad Chitro: A Multimodal Bangla News Image-Text Dataset contains 50,000 Bangla news articles and 50,000 associated image files. It supports research in Bangla natural language processing, computer vision, and multimodal learning. The articles were collected from Prothom Alo: 49,926 from http://www.prothomalo.com and 74 from affiliated Prothom Alo platforms (Bigganchinta, Nagorik, Haal Fashion, Kishor Alo and Bondhushava). The recorded publication dates range from 4 December 2025 to 30 August 2026. The articles are organized into nine major categories: বাংলাদেশ (Bangladesh), রাজনীতি (Politics), বিশ্ব (World), বাণিজ্য (Business), মতামত (Opinion), খেলা (Sports), বিনোদন (Entertainment), চাকরি (Jobs), and জীবনযাপন (Lifestyle). The dataset preserves 113 original subcategory labels, providing a two-level structure for broad and detailed news classification. Each article record includes its title, body text, subcategory, major category, publication date, source URL, author, image caption where available, image URL, and local image path. These fields connect textual content with associated images and source information. Editorial image captions are available for 45,095 articles (90.19%). Major categories were assigned using the source sections recorded in article URLs, supplemented by custom mappings for topics outside the nine-category structure. Education is grouped under Bangladesh, literature under Entertainment, religion and general science and technology under Lifestyle, and freelancing under Jobs. A Category Mapping worksheet documents these groupings and their article counts. The dataset is distributed as an Excel workbook (Shongbad Chitro.xlsx) and an images folder. The workbook contains three worksheets: articles, Summary, and Category Mapping. The articles worksheet also retains 86 empty trailing rows, which are excluded from the total of 50,000 articles. All 50,000 image files decode correctly. 49,998 are JPEG and 2 are BMP, although some files carry .avif, .png or .webp extensions. Some images are generic illustrations that the newspaper reuses across articles: 9,721 files (19.44%) are exact copies of another image file. Shongbad Chitro can support research on Bangla news classification, hierarchical classification, multimodal topic prediction, image–text retrieval, image captioning, and news summarization. The CC BY 4.0 licence covers only the authors' contribution (the compilation, record structure, category mapping and derived metadata). It does not grant any rights in the underlying news text, images or captions, which remain the property of Prothom Alo and the credited authors and photographers.
Files
Institutions
- International Islamic University ChittagongChittagong, Chittagong