Learning the Algorithm: YouTube Creator Discourse Dataset

Learning the Algorithm: YouTube Creator Discourse Dataset

This dataset contains transcripts of 179 YouTube videos in which content creators discuss the YouTube algorithm, platform visibility, and optimization strategies. The corpus was collected and analyzed for the paper "Learning the Algorithm: Responsibilization and the Limits of Agency on YouTube" (Urban, 2026, submitted to New Media & Society).

The dataset supports research on algorithmic folk theories, platform governance, creator labor, and computational discourse analysis.

Dataset Contents

youtube_algorithm.csv — Main dataset containing video metadata and transcripts for 179 videos, including video ID, title, publication date, matched search term, channel ID, view count, like count, comment count, duration, subscriber count, total channel views, total videos, detected language, and full transcript. Data collected via YouTube Data API v3.

codebook.csv — Thematic framework documenting the 8 algorithmic folk theory themes identified through iterative qualitative analysis, including conceptual definitions and seed quotes used as semantic anchors for classification.

Collection Method

Videos were identified using targeted search queries including "YouTube algorithm," "how the YouTube algorithm works," "beat the YouTube algorithm," "why my views dropped," and "grow on YouTube." Videos were included if they explicitly addressed YouTube's algorithmic systems in English and had transcripts of at least 40 words.

Replication

To replicate the analysis: load youtube_algorithm.csv for transcript data and codebook.csv for the thematic framework. Apply semantic similarity classification using seed quotes as anchors with the all-mpnet-base-v2 sentence embedding model (cosine similarity threshold: 0.45). Full methodological details are provided in the paper.

License

Released under CC0 1.0 Universal (public domain dedication). Note: individual videos remain copyrighted by their creators. This license applies to the compiled dataset and metadata only.

Contact

Nadia Urban, East China Normal University

ORCID: 0009-0001-6282-5919