Uzbek Sentiment + Named Entity Recognition Dataset

This dataset contains 12,000 Uzbek (Latin script) sentences annotated for sentiment (positive/negative/neutral) and named entities (PER, ORG, LOC). Each record includes the raw sentence text, a numeric polarity score (-1/0/1) with the corresponding sentiment label, and a list of entity mentions aligned with their entity types. The dataset additionally provides emoji-related signals: extracted emojis, their position within the sentence, emoji sentiment, and a conflict flag indicating disagreement between text polarity and emoji polarity. Overall, the corpus includes 24,208 entity mentions and 4,548 emoji tokens (14 unique emojis). The dataset is intended for training and evaluation of sentiment classification, NER, and integrated pipelines such as “Sentiment → NER”, especially for low-resource Uzbek NLP and user-generated content analysis.