Polarization by recommendation: analyzing YouTube’s polarization dynamics around Dutch political parties

Given previous findings regarding YouTube’s pathways toward radicalization, this research investigated to what extent YouTube’s recommendation algorithm contributes to the affective polarization of Dutch watchers of political parties’ content. To investigate this, content networks around parties’ YouTube presence were generated and video subtitles and titles were extracted for analysis. Sentiment classifiers were trained to recognize the presence of polarization and negative sentiment. Statistical analyses were then employed to test for differences between different political parties’ content networks. Results indicate that the negativity of a video title is positively associated with its polarization level, and that videos from right-wing parties produce more polarizing recommendations. This analysis reveals the recommendation algorithm’s role in the spread of for increasingly polarizing content. Currently, it contributes to the increased spread of polarizing content, especially for users who are watching content within right-wing party networks.