10.48788/DVUA/BM3ACVZharkov, DmytroDmytroZharkov0009-0006-2700-5313V.M. Glushkov Institute of Cybernetics of the NAS of UkraineDataverseUA2026Computer and Information Sciencetext classificationmachine learningUkrainian languageNLPdocument classificationlarge language modelsZharkov, DmytroDmytroZharkovV.M. Glushkov Institute of Cybernetics of the NAS of UkraineV.M. Glushkov Institute of Cybernetics of the NAS of UkraineNational Academy of Sciences of Ukraine2026-04-242026-08-25269930245019934text/plainapplication/jsonapplication/json1.0CC BY 4.0The dataset contains 355 classified advertisements organized into 15 semantic categories and represented as structured JSON objects for supervised multi-class text classification. Each advertisement includes a unique identifier, category identifier and title, advertisement title, full advertisement text, and an LLM-assisted summary. The accompanying category data provide category identifiers and titles together with category-level bag-of-words (BOW) and TF-IDF representations derived from the advertisement corpus. The corpus consists predominantly of Ukrainian-language advertisements and includes naturally occurring mixed Ukrainian–Russian content. The texts preserve characteristics of real-world advertisements, including spelling variations, colloquial language, repetitions, commercial information, and stylistic variability. The dataset covers multiple thematic domains, including furniture, commercial premises and rentals, cosmetics, perfumery, healthcare and beauty products and services, medical products, and equipment. The dataset is intended for research and educational purposes and can be used for supervised text classification, evaluation and benchmarking of machine learning and large language model (LLM)-based classifiers, natural language processing research, feature engineering, and comparative evaluation of text classification methods.The corpus comprises 355 classified advertisement records distributed across 15 semantic categories. The textual content is predominantly Ukrainian and includes mixed Ukrainian–Russian language features characteristic of real-world user-generated advertisements. Records preserve spelling variations, colloquial expressions, commercial information, repetitions, and stylistic variability. Each advertisement is associated with a single category label.The dataset is accompanied by a README file describing the dataset purpose, JSON file structure, advertisement and category objects, category labeling, dataset characteristics, and recommended uses. The dataset includes adverts.json with 355 structured advertisement records and categories.json with 15 category definitions and corresponding bag-of-words (BOW) and TF-IDF representations.