Observatory
YOUPOL / ExploreThe corpus over time
Content measures are computed from videos that have been transcribed and analysed down to the sentence. Periods with too little coverage are not plotted.
Videos published per year
How to read these charts
Videos of the tracked channels, by year of publication. Transcribed videos are broken down by orientation. The hatched part groups the videos tracked but not yet transcribed; it is larger for recent years, which are still being transcribed.
Videos transcribed and awaiting transcription
In 2026, 14% of tracked videos are transcribed and analysed down to the sentence.
Share of political content
How to read these charts
Each transcript sentence and each comment sentence is classified as political or not. The first chart follows the whole corpus; the next two compare orientations. All three are plotted by quarter of publication and count YouTube and TikTok videos together.
Political content over time
Share of transcript and comment sentences classified as political, by quarter of publication of the videos, with the volume analysed. A quarter is plotted when it has at least 10,000 sentences.
Annotation and validation
Training sentences are annotated by a language model. For the politicisation of transcripts, its reliability was first measured on 1,000 sentences also annotated, independently, by two researchers. This human check does not cover comments. Each classifier is then evaluated on a validation set kept apart from training.
| Measure | Value |
|---|---|
| Human–model agreementLight’s κ, 1,000 sentences | 0.787 |
| Macro F1language model against human consensus | 0.922 |
| Transcript classifierMacro F1, 2,000 validation sentences | 0.92 |
| Comment classifierMacro F1, 4,339 validation sentences | 0.89 |
Sources: Lemor and Boursier, YOUPOL project paper, table 2 (agreement and F1 of the language model); the project’s model registry (classifiers).
Tool LLM_Tool (technical paper)
Nine themes measured
How to read these charts
Weight of each of the nine themes in a typical video, by orientation and year of publication: mean, over the videos, of the share of a video’s political sentences dealing with the theme.
The classifiers detect the theme of a sentence, whatever the position expressed. They were trained on annotations designed to measure far-right and neo-reactionary ideas, but each category was defined by its subject: a criticism of migration policy counts as a sentence about immigration.
- Immigration
- Political sentences about immigration: foreigners, immigrants, migrants, asylum.
- Democracy
- Political sentences about democracy, as an ideal or as a regime: elections, institutions, representation.
- Authority
- Political sentences about authority: obedience, order, the use of force against those who break norms.
- Tradition
- Political sentences about tradition: values, customs, family models, heritage.
- Progress
- Political sentences about progress in a broad sense: technical, human or social progress.
- Equality and inequality
- Political sentences about equality or inequality between human beings: sexes, origins, social groups.
- Environment
- Political sentences about the environment or ecology: climate, energy, nature, green parties.
- Technology
- Political sentences about technology, technique or innovation, or about tech companies and their leaders.
- Distrust of the state
- Political sentences expressing libertarian ideas or distrust of the state, public services, taxation or norms.
Public figures named by each orientation
How to read these charts
Public figures most often named in the videos of each orientation, all years combined. The rate is the number of sentences naming them per 10,000 transcribed sentences of that orientation, which adjusts for differences in corpus size.
The ten most frequently named public figures
Per 10,000 transcribed sentences of the selected orientation. Only orientations with at least 10,000 transcribed sentences are shown.
Politicisation of videos and their comments
How to read these charts
Share of comment sentences classified as political, by the share of political sentences in the video. Only videos with at least 200 sentences and at least 50 comment sentences are included.
Comment politicisation by video politicisation
Mean per 10-percentage-point bin; bins with fewer than 10 videos are omitted. Orientations with at least 15 videos.
Correlation by orientation
Pearson correlation between a video’s share of political sentences and that of its comments. Orientations with at least 15 videos.
| Orientation | r | Videos |
|---|---|---|
| Far right | 0.19 | 93 |
The formats of the videos
How to read these charts
Extracted YouTube videos, by orientation and country: length, share of short and long videos, share of monologues (one speaker accounts for at least 90% of the text) and share of political sentences. Rows with fewer than 30 videos are not shown. A dash marks too small a base: fewer than 30 transcribed videos or fewer than 2,000 sentences.
The two countries should be compared within the same orientation: the Quebec corpus has no left-wing channel.
| Orientation | Country | Extracted videos | Average length | < 3 min | ≥ 60 min | Monologues | Political sentences |
|---|---|---|---|---|---|---|---|
| Far right | Quebec | 5,163 | 21.3 min | 34.2% | 7.3% | 41.0% | 46.2% |
| Manosphere | Quebec | 790 | 8.4 min | 67.8% | 2.8% | 82.2% | 1.9% |
| Conspiracism | Quebec | 418 | 13.3 min | 8.1% | 3.3% | 76.4% | — |
Videos that became unavailable
How to read these charts
Tracked videos that later became unavailable, by year of publication: removed by the platform or by the author, made private, or channel closed. These videos are re-checked every month.
Unavailable videos by year of publication
Latest video: publication date of the most recent tracked video. Observed on: date on which the project recorded the closure.
| Channel | Orientation | Latest video | Observed on | Reason given |
|---|---|---|---|---|
| Gabriel Duquette | Manosphere | 13 March 2024 | 20 March 2026 | account deactivated |