HongguoTrending: A Longitudinal, Multimodal Archive of Trending Videos and Topics on a Leading Micro-Drama Platform

Rongxin OuyangarXiv 2026 · arXiv

TL;DR

Researchers have released HongguoTrending, the first longitudinal, multimodal dataset tracking the "micro-drama" (duanju) phenomenon on China’s leading platform, Hongguo. By archiving hourly snapshots of trending lists alongside video samples and engagement metrics, the paper reveals how algorithmic gatekeeping, kinship-centered narratives, and production-side "mainstream values" shape this multi-billion dollar digital frontier.

What Problem Does This Paper Address?

Micro-dramas—high-volume, serialized vertical videos—have exploded in popularity, with the Chinese market alone generating over 50 billion yuan in 2024, surpassing the national film box office. Despite this massive cultural shift, researchers have lacked the data to understand visibility.

Existing datasets focus on video quality or content understanding but are typically static snapshots. They don't show how a series climbs or falls on a trending list over time. This paper addresses the "missing link" between multimodal content (what the video looks like) and platform visibility (how the algorithm promotes it), allowing researchers to study algorithmic gatekeeping and popularity dynamics in real-time.

Methods

The authors developed a two-stage asynchronous pipeline to scrape Hongguo (a ByteDance platform) on an hourly schedule:

  1. Metadata Worker: Extracts series attributes (IDs, titles, search scores, cast, and genre tags) from seven different trending lists across web and mobile interfaces.
  2. Asset Scheduler: Dispatches workers to fetch "heavy" multimodal data, including cover images and the first five episodes of each series.

The data is stored in a PostgreSQL database with assets on S3-compatible storage. To analyze the data, the authors employed:

  • NLP: Using the Jieba library for Chinese text segmentation to analyze synopses and titles.
  • Computer Vision: Utilizing YOLO for object detection and Apple Vision OCR to map the spatial layout of text on series covers.
  • Econometrics: Applying two-way fixed-effects models to determine which engagement metrics (likes, shares, etc.) actually predict changes in trending rank.

Datasets

The primary contribution is the HongguoTrending dataset itself. As of the initial snapshot, it includes:

  • 63,200 ranking entries across 158 hours.
  • 889 distinct series (categorized into live-action, AI-generated, and animated).
  • 2,689 video episodes (totaling ~24.8 GB).
  • 1,007 cover images.
  • Audience metrics: Hourly logs of "heat" scores, ratings, likes, favorites, and comments.

The dataset is accessible via a RESTful API for authorized academic use.

Key Insights and Ideas

  • The "Mainstream" Supply Push: A striking 83% of series conform to "core socialist values." However, these series don't actually receive higher engagement than non-conforming ones. This suggests that the prevalence of these themes is driven by production supply and platform curation rather than organic audience demand.
  • Kinship as a Structural Core: While titles often use "high-concept" hooks (like rebirth or apocalypse), the synopses are overwhelmingly centered on family. Terms like mother, husband, and daughter form the dense core of the narrative network, suggesting these dramas are essentially modernized, hyper-compressed family soaps.
  • Sharing is the "Strong Signal": In the race for visibility, not all engagement is equal. The authors found that user sharing is the strongest predictor of a series gaining rank the next day. Passive "likes" or "favorites" had a much weaker correlation with future visibility.
  • Visual Grammar: Micro-drama posters follow a rigid "visual template": character portraits in the top half and prominent title typography in the bottom half (covering about 20% of the total area).

Key Results

  • Low Interface Overlap: There is surprisingly little overlap between what trends on the Hongguo website versus the mobile app (a median Jaccard similarity of only .084). The mobile app turns over much faster, suggesting different algorithmic priorities for different devices.
  • Rapid Turnover: On the web's "overall" list, half of the new entries disappear within just 3 hours.
  • Predictive Power: A one-standard-deviation increase in shares predicts a gain of 8.037 positions on the client trending list over 24 hours.
  • Color Palettes: Cover colors are highly genre-specific. For example, "rural" and "period" dramas use significantly more orange (nostalgic tones), while "urban" dramas lean into black and "otherworld" dramas into blue.

Limitations and Future Work

The authors acknowledge several constraints:

  • Single Platform: The data comes only from Hongguo; findings might not generalize to international apps like ReelShort.
  • Lack of Personalization: The data was collected via anonymous profiles, so it doesn't account for how the algorithm personalizes lists for individual users.
  • Short Window: The initial analysis covers a relatively short time frame, missing seasonal trends.
  • Correlational Nature: The relationship between engagement and heat is correlational; the exact proprietary "heat" formula remains a black box.

Future work involves expanding the archive to include audio transcripts and using the panel structure for quasi-experimental evaluations of platform policy changes.

Summary updated

Abstract

Micro-dramas (duanju) are reshaping digital production and consumption, yet a lack of longitudinal data connecting series to their platform visibility persists, limiting our understanding of the emerging medium. We introduce HongguoTrending, a longitudinal, multimodal dataset of hourly snapshots from seven trending lists of Hongguo, China's largest micro-drama platform. Collected through a two-stage asynchronous pipeline, the archive links 63,200 ranking entries across 632 web list-hours and 157 client hours (889 distinct series) to metadata, cover images, sampled episode videos, and audience engagement metrics. Exploratory analyses show distinct turnover rates between web and mobile trending lists, production-driven mainstream values alongside kinship-centered synopses, genre-specific cover palettes, and strong associations between user sharing and next-day rank gains. Accessible via a RESTful API, this dataset supports empirical research on algorithmic gatekeeping, popularity dynamics, and multimodal digital culture.
BibTeX
@misc{ouyang2026hongguotrending,
  author = {Rongxin Ouyang},
  title = {{HongguoTrending: A Longitudinal, Multimodal Archive of Trending Videos and Topics on a Leading Micro-Drama Platform}},
  year = {2026},
  eprint = {2611.19351},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.19351},
}
Figure