HyperBulAds: Dataset Of Hyperboles in Bulgarian Health and Beauty Advertising Texts

HyperBulAds is a pilot dataset designed to support research on hyperbolic language in Bulgarian advertising texts, specifically within the domain of health and beauty product descriptions. It consists of 623 manually annotated sentences, each labeled for the presence of hyperbole and, when applicable, annotated at the span level with one of three categories: qualitative quantitative generalization This dataset enables both binary classification (1-hyperbolic vs. 0-non-hyperbolic) and sequence labeling (identifying hyperbolic segments). Example (non-hyperbolic): { "text": "Чаят има множество положителни действия върху организма.", "hyperbole": false, "spans": [] } Example (hyperbolic): { "text": "100% натурален супербленд за здрав стомах и тънка талия.", "hyperbole": true, "spans": [ { "start": 0, "end": 14, "label": "quantitative", "text": "100% натурален" } ] } Annotation Methodology All annotations were conducted manually by a single expert linguist. As this is an exploratory pilot study, the process was carried out without formal guidelines or inter-annotator validation—an acknowledged limitation that will be addressed in future iterations. Dataset Statistics Total sentences: 623 Hyperbolic sentences: 263 Non-hyperbolic sentences: 360 Distribution of hyperbole types: Qualitative: 208 instances Quantitative: 42 instances Generalization: 36 instances This distribution illustrates the prevalence of qualitative exaggeration in promotional texts for health and beauty products. The dataset is provided in JSON format, suitable for direct use in NLP pipelines and machine learning experiments. All content has been anonymized, with brand and product names removed to protect privacy and minimize commercial bias. Access HyperBulAds is with open-access to support transparency, reproducibility, and collaborative research. https://doi.org/10.5281/zenodo.15723330 Authors Nevena Grigorova, GATE Institute Silvia Gargova, GATE Institute Funding This resource is published with the assistance of the ORBIT project (Open Research Bulgarian Incubator), a cascading initiative supported by the REINFORCING project, funded by the European Commission. Citation Please cite this dataset as: Gargova, S., & Grigorova, N. (2025). HyperBulAds: Dataset Of Hyperboles in Bulgarian Health and Beauty Advertising Texts [Data set]. Zenodo. https://doi.org/10.5281/zenodo.15723330