BNHH-Context: A Manually Curated Synthetic Bengali Text Dataset for Context-Aware Halal and Haram Food Classification

BNHH-Context is a manually curated Bengali text dataset developed for context-aware binary classification of food-related statements as Halal or Haram. The dataset contains 3,150 unique Bengali text records, including 1,704 Halal records (54.1%) and 1,446 Haram records (45.9%). Each record contains structured contextual information, including a unique sample identifier, Bengali text, food entity, binary label, category, rule identifier, contextual condition, ingredient/evidence information, group identifier, scenario identifier, source-reference URL, review status, and construction provenance. All sample identifiers, sentence texts, and scenario identifiers are unique. The dataset was developed using manually designed food entities and contextual scenarios. Bengali sentences were manually authored according to predefined rule families grounded in Quranic references (2:168, 2:173, 5:3, 5:90, 5:96, and 6:145) and Sahih Muslim 2003a for intoxicant-related conditions. The dataset covers plant-based foods, packaged and mixed foods, cooked-food ingredient contexts, prohibited additions, fish preparation, slaughter and carrion contexts, and additional food-related categories. BNHH-Context is intended to support research in Bengali natural language processing, contextual text classification, rule-aware modeling, leakage-aware evaluation, error analysis, and interpretable halal-food information systems. The contextual design allows the classification of a food-related statement to change according to explicitly stated ingredients, conditions, or rule-relevant evidence rather than relying only on the food name. The dataset is provided as a research resource for computational modeling and data-method development. Its labels represent the rule framework used during dataset construction and should not be interpreted as formal halal certification, religious rulings, or product-specific expert decisions. No human participants, identifiable personal information, surveys, interviews, clinical records, sensor-derived data, field-site data, or animal experiments were involved in the construction of this dataset.