Year: 2025
Type: conference-paper
Abstract: Short video platforms have garnered significant attention in recent years, with discussions ranging from concerns about inappropriate content and addiction to strategies for maximizing user engagement and screen time. Despite the large user base and growing relevance of these platforms, there is still a notable lack of comprehensive datasets focused on broad recommendation and moderation. This is especially true for TikTok, where API access is limited and collecting unbiased data is challenging. In this collection, we present a diverse and rich dataset from YouTube Shorts and TikTok’s main feeds in Brazil, comprising over 35,000 videos. The dataset includes detailed engagement statistics, extensive video metadata, over 170,000 keyframes for visual analysis, and Safe Search API assessments for each keyframe. This rich resource fills a critical data gap, offering valuable tools for research on content categorization, user behavior analysis, and platform engagement strategies.
Authors: Tomas Lacerda, Marcelo Sartori Locatelli, Igor Costa, Lorenzo Carneiro, Virgílio Almeida, Wagner Meira Jr.
Institution: Universidade Federal de Minas Gerais
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4413791635