Abstract
Bangla REX introduces a significant dataset designed specifically for relation extraction tasks in the Bangla language, addressing a critical lack of resources in this domain. To the best of our knowledge, this dataset fills an important gap by providing comprehensive support for advanced relation extraction methodologies. The dataset comprises 90,441 text entries, available in structured formats. Alongside this, a Bangla Knowledge Base (KB) containing 63,256 entries has been developed to automate the corpus annotation with relation tags. The dataset is developed to support the accurate extraction and classification of various semantic relations within Bangla textual data. It includes categories such as movie actors, place of birth, movie directors, writers, places of death, company locations, and founders. Bangla REX is a valuable resource for researchers and practitioners in natural language processing, offering standardized benchmarks and detailed annotations to advance relation extraction techniques for Bangla. It aims to foster research and enable practical applications, contributing to the development of Bangla NLP and multilingual information processing. The dataset is compiled from reliable sources, including Bangla Wikipedia and Wikidata, and has been annotated to ensure accuracy and relevance. Each relation category contains numerous labeled examples, providing a strong foundation for training and evaluating relation extraction models. Bangla REX seeks to promote innovation in NLP research and support the development of tools tailored to the Bangla language.