Patterns in Polarising YouTube Content Recommendations Following Dutch Political Party Videos

The final speaker in this ECREA 2024 session is João Gonçalves, whose interest is in polarisation in the content recommended by YouTube in the Netherlands. This focusses especially on affective polarisation, on labelling of out-groups as extremist, and on a lack of discursive crossover between polarised opponents.

Past research on polarisation has shown a substantial role for non-news Websites; investigating YouTube recommendations is therefore especially important. A key distinction introduced in the present study is between content recommendations around established and non-established parties; additionally, the study also explored content recommendations specifically around right-wing parties.

The project selected five seed videos from each of the selected parties, and then captured the additional videos recommended alongside those seed videos. It classified these videos as polarising or non-polarising by applying a text classifier to the captions of each video. The results showed that content from non-established parties did not produce any more polarised recommendations, but that right-wing content produced more polarised recommendations. Also, more polarised seed videos produced more polarised further video recommendations.

In other words, content recommendations around right-wing parties, not non-establishment parties, drive polarised content recommendations; similarly, already polarised videos drive the recommendation of additional polarised videos. This suggests that the recommender system is picking up on cues in the seed videos, and matches these to cues in the videos it recommends.