Loading...
Loading...
News corpora, social media analysis, movie/music metadata, sports data, cultural datasets, misinformation
11,243 datasets
Six cultural variants of the GSM8K test set were created to investigate how cultural context influences large language model performance on mathematical problems. The dataset is part of a research project detailed in the paper 'Lost in Cultural Translation: Do LLMs Struggle with Math Across Cultural Contexts?' and was authored by abedk. It was last updated on Hugging Face in March 2025.
KidzRizal published a dataset of Twitter posts related to the Indonesian President on Hugging Face on 2025-05-13. The dataset likely contains text data for sentiment analysis. Metadata is minimal; actual content requires verification after download.
DurLAR is a high-fidelity 128-channel LiDAR dataset. It includes an exemplar subset of 600 frames available for preview, featuring image-based topics such as image_00, image_01, reflectivity, and ambient data. The dataset is intended for autonomous driving and robotics applications.
Ivano-Frankivsk Regional State Administration provides a summary report on the activities of club establishments under its Department of Culture, Nationalities and Religions. The dataset was last updated on 2025-04-01 and is available in common spreadsheet formats. It originates from the States site of Ukraine, a national open data platform.
A text dataset containing summaries of shipping news articles, authored by Ktzoras. The dataset was last updated on HuggingFace on May 8, 2025. The specific number of articles, summary length, and source publications are unknown from the provided metadata.
A Vietnamese multilabel dataset designed to assess large language models in verifying electronic information for legal contexts. The dataset, named RMDM, contains four labels: real, misinformation, disinformation, and mal-information. It was created by nguyenthanhasia and last updated on March 31, 2025.
DeepReviewer-13K captures a human-like deep thinking process for peer review. The dataset contains structured annotations and intermediate reasoning steps, curated by WestlakeNLP. It was last updated on March 18, 2025.
MovieTection is a benchmark dataset containing between 10,000 and 100,000 records designed to detect copyrighted visual content in Large Vision-Language Models (VLMs). Developed by the DIS-CO research team in early 2025, it provides image-caption pairs to evaluate model exposure to pretraining data. The dataset supports Membership Inference Attacks (MIA) by testing if models can identify specific movie titles from frames or text descriptions.
312 apps across 33 categories from 6 major domains including Entertainment, Productivity, and Health. It contains over 115 hours of exploration videos for context-aware learning, created by author 9211sun and last updated on March 21, 2025.
Over 120,000 reviews for research papers crawled from the OpenReview platform, covering conferences such as ICML, NeurIPS, ICLR, and CVPR. The dataset includes paper metadata, review content, and rating information. It was uploaded by author guochenmeinian and last updated on March 17, 2025.
MixCuBe (Mixed Culture Benchmark) accompanies a 2025 paper evaluating cultural awareness in multimodal large language models. The dataset likely contains mixed-cultural inputs combining elements from multiple cultures, such as images and text. It was created by kyawyethu and published on Hugging Face.
HateBR is the first large-scale, expert-annotated dataset of Brazilian Instagram comments for hate speech detection. It contains 7,000 documents manually annotated by specialists, collected from comments made by politicians. The dataset was last updated on March 9, 2025.
1,000 synthetic political news headlines labeled with perceived political bias (Left, Center, Right). The dataset was created by aryn25 for the BiasLens project to train LLMs and classifiers for detecting political bias in media. It was last updated on March 26, 2025.
The Triple-to-Text Alignment dataset aligns Knowledge Graph triples from Wikidata with diverse, real-world textual sources extracted from the web. It provides a broader range of writing styles, tones, and structures by leveraging Wikidata references from sources such as news articles, government reports, and scientific publications. The dataset was created by sven-h and was last updated on 2025-03-15.
CSL-News is a large-scale Chinese Sign Language dataset proposed in the paper 'Uni-Sign: Toward Unified Sign Language Understanding at Scale'. It is designed for developing robust sign language understanding models and is hosted on HuggingFace by author ZechengLi19. The dataset was last updated on March 14, 2025.
Lao News classification dataset collected from laopost.com. The dataset was created by author wannaphong and was last updated on March 4, 2025. It is structured into train, validation, and test splits, with each article containing a title field.
Zeta is an open dataset used to train Zeta, an edit prediction model for Zed's predictive coding feature. The dataset is derived from Qwen2.5-Coder-7B and is designed to predict a developer's next code edit based on recent programming patterns and cursor position. It was published by zed-industries and last updated on February 27, 2025.
Multilingual customer reviews from the Amazon platform, aggregated by the user 'buruzaemon' and hosted on Hugging Face. The dataset was last updated on March 26, 2025. The specific source, scale, and content details are described on the linked Hugging Face page.
Saarland Landscape Program data defines green fences for cultural landscape and recreation planning. Bundesamt für Kartographie und Geodäsie provides this spatial data via an OGC API Features service. The dataset was last updated on February 25, 2025.
897,000 images of wine labels and 824,000 wine reviews curated from the Vivino platform. The dataset includes over 350,000 unique vintages annotated with year, region, rating, alcohol percentage, price, and grape composition. A subset includes fine-grained flavor annotations from a tasting experiment with 256 participants.