Reproduce “A Large-scale Dataset with Behavior, Attributes, and Content of Mobile Short-video Platform” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yu Shang, Chen Gao, Nian Li, Yong Li

A new large-scale dataset, derived from a real mobile short-video platform, offers 1,019,568 interactions from 10,000 users with 153,561 videos, encompassing diverse user attributes, a hierarchical video categorization system, and 3.2 TB of raw video content. This resource supports advanced research in user modeling, computational social science, and multimodal recommendation, demonstrating the efficacy of multimodal algorithms where BM3 outperformed others across various metrics, and revealing differentiated filter bubble effects for active versus inactive users.

Key insights

  • The dataset's integration of rich behavioral data, diverse attributes, and raw multimodal content effectively addresses critical data gaps in previous short-video platform datasets, enabling broader research beyond traditional recommendation tasks.
  • Multimodal features derived from video content are crucial for enhancing recommendation algorithm performance, with specialized multimodal models generally outperforming general methods.
  • User engagement levels play a role in susceptibility to filter bubbles; active users tend to maintain content diversity more effectively than inactive users, whose content exposure becomes narrower over time.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.