Conceptual
Browse Instruct Curate

Contextual Feature-Fusion Headline Generation for Bengali Religious News

Introduces the BeliN corpus of Bengali religious news labeled with category, aspect, and sentiment, and MultiGen, which fuses these contextual features with the article body as multi-input to transformer pretrained language models (BanglaT5, mBART, mT5, mT0), improving abstractive headline generation over content-only baselines and demonstrating the value of contextual features for low-resource-language summarization.