Twits, Toxic Tweets, and Tribal Tendencies: Trends in Politically Polarized Posts on Twitter

Hans W. A. Hanley,Zakir Durumeric
2024-10-31
Abstract:Social media platforms are often blamed for exacerbating political polarization and worsening public dialogue. Many claim that hyperpartisan users post pernicious content, slanted to their political views, inciting contentious and toxic conversations. However, what factors are actually associated with increased online toxicity and negative interactions? In this work, we explore the role that partisanship and affective polarization play in contributing to toxicity both on an individual user level and a topic level on Twitter/X. To do this, we train and open-source a DeBERTa-based toxicity detector with a contrastive objective that outperforms the Google Jigsaw Perspective Toxicity detector on the Civil Comments test dataset. Then, after collecting 89.6 million tweets from 43,151 Twitter/X users, we determine how several account-level characteristics, including partisanship along the US left-right political spectrum and account age, predict how often users post toxic content. Fitting a Generalized Additive Model to our data, we find that the diversity of views and the toxicity of the other accounts with which that user engages has a more marked effect on their own toxicity. Namely, toxic comments are correlated with users who engage with a wider array of political views. Performing topic analysis on the toxic content posted by these accounts using the large language model MPNet and a version of the DP-Means clustering algorithm, we find similar behavior across 5,288 individual topics, with users becoming more toxic as they engage with a wider diversity of politically charged topics.
Social and Information Networks,Computers and Society
What problem does this paper attempt to address?
### What problems does this paper attempt to solve? This paper aims to explore and quantify the relationship between political polarization and toxic content on the social media platform Twitter. Specifically, the author attempts to answer the following key questions: 1. **The relationship between partisanship, political polarization and users' posting of toxic content**: - The author has studied the partisanship of politically - engaged users, political polarization, and how these factors influence their tendency to post toxic content. - The paper has analyzed whether users' political stances are related to the frequency of their posting of toxic content, and how users' interactions with others of different political views affect their toxic behaviors. 2. **The influence of user characteristics on toxic topics**: - Researchers have analyzed how user characteristics, including partisanship, predict their toxic topics on Twitter by collecting 89.6 million tweets. - Specifically, the author uses the Generalized Additive Model (GAM) to determine the relationship between user characteristics (such as partisanship, account age, etc.) and the frequency of posting toxic content. 3. **Political polarization and toxicity at the topic level**: - The author further explores how political polarization affects the generation of toxic content at the specific topic level. - By using the large - language model MPNet and the DP - Means clustering algorithm to analyze 5,288 independent topics, the research finds that when users are exposed to a wider range of political topics, the toxic content they post will increase. ### Method overview To answer the above questions, the author has adopted the following methods: - **Data collection**: 89.6 million tweets were collected from 43,151 Twitter users. - **Toxicity detection model**: A contrastive - learning - based toxicity detection model based on DeBERTa was trained and open - sourced. This model outperforms the Google Jigsaw Perspective Toxicity detector on the Civil Comments test set. - **User partisanship estimation**: Correspondence Analysis (CA) was used to estimate users' partisanship according to the political figures they follow. - **Topic analysis**: The MPNet and DP - Means clustering algorithms were used to conduct topic analysis on toxic tweets to identify the topics with the most toxic and polarized characteristics. ### Main findings - The more users interact with toxic content and users of other different political views, the higher the probability that they will post toxic content themselves. - Users' political stances themselves have little influence on the frequency of their posting of toxic content, but are closely related to the diversity of their interactions. - At the topic level, as users participate in more diverse political topics, the toxicity of their tweets will also increase, showing strong tribal tendencies and emotional polarization. In conclusion, this study reveals the complex relationship between political polarization and toxic content and provides valuable insights for understanding negative interactions in online communities.