Our method · 01

Detect

The channels’ feeds every six hours, their full list every week, the last seven days first.

  1. 1Each channel’s feeds are read every six hours.
  2. 2Every week, the full list of each channel’s videos is reread.
  3. 3A detected video enters the queue with its publication date.
  4. 4The videos of the last seven days go ahead of everything else.
Collection

Detection, extraction and audience snapshots

Collection, transcription and sentence classification are carried out by the project’s network of computers. The central server hosts the database and this site. It also computes the vectors used for meaning-based search and writes the automatic texts.

  1. every 6 h

    Detection, every six hours

    The public RSS feeds of each YouTube channel (long videos, Shorts, live streams) give new videos, their exact publication date and their view count.

  2. every week

    Full listing, every week

    A full listing of each channel’s videos is taken every week. It catches videos missing from the feeds and reveals deletions. TikTok accounts, which have no RSS feed, are covered by this listing only.

  3. within the hour

    Metadata and comments

    Within an hour of detection, a video less than nine days old moves to the front of the collection queue: metadata, first view count and comments. Older videos are extracted in batches.

  4. day 1 · 7 · 30 · monthly

    Audience figures

    Views, likes and comment counts of recent videos are measured again after one day, one week and one month. The whole corpus is also re-measured once a month. Comments on recent videos are extracted again after one week and one month.

  5. last 7 days first

    Transcription

    Videos published in the last seven days always come first, whatever their length. Next come all videos published since 1 April 2026, then the batches chosen by the team and the rest of the backlog. A video detected but not yet extracted is transcribed directly.

Terms used on this site
Tracked
Any video recorded in the database, whatever its status.
Detected
Known from a feed or a listing, not yet extracted.
Extracted
Metadata and comments extracted.
Transcribed
Speech converted into text.
Analysed
Transcript split into sentences, which the classifiers have labelled.
Tracked channel
Channel whose new videos YOUPOL detects and collects.
Project paper