BanTABSA: A Temporally and Geographically Annotated Bangla Aspect-Based Sentiment Analysis Dataset for Hotel Reviews
Description
BanTABSA is a manually annotated dataset of Bengali hotel reviews designed for Aspect-Based Sentiment Analysis (ABSA), fine-grained opinion mining, and spatio-temporal NLP research. The dataset comprises 1,242 reviews yielding 3,633 aspect-opinion-sentiment triplets collected across 31 hotel establishments in 9 districts of Bangladesh, aggregated from Facebook groups, TripAdvisor, Booking.com, and guest surveys spanning review dates from 2017 to 2026. Key features of BanTABSA: • Fine-Grained Annotations: Contains aspect-opinion-sentiment triplets across 24 aspect categories and 3 sentiment polarities (positive, negative, neutral). • Geo-Temporal Grounding: Each review is paired with a standardized ISO-8601 timestamp and hotel/district location identifiers, enabling temporal decay and geographical sentiment studies. • Benchmark Evaluation: Accompanied by baseline classification evaluations across TF-IDF, FastText, IndicBERTv2, XLM-RoBERTa, and BanglaBERT models. Files included in this release: 1. BanTABSA.csv: Primary dataset containing all reviews, ratings, timestamps, location metadata, and JSON-formatted triplets. 2. README.md: Data schema, column descriptions, taxonomy details, and Python usage code. 3. survey_questionnaire.pdf: The guest survey instrument used during data collection.
Files
Institutions
- Khulna University of Engineering and TechnologyKhulna Division, Khulna