Echoes on the Internet: Dissecting Social Media Silos through Behavioral and Personality Markers

In social media, a user can interact with other users who hold similar viewpoints, contributing to the creation of “echo chambers”. Echo chambers are networks where users’ viewpoints are not only highly homogenous but also mutually reinforcing; echo chambers can intensify societal divisions and facilitate the spread of misinformation. Therefore, it is crucial to identify the risk features that make users susceptible to joining echo chambers. In this context, this study pioneered the application of machine learning methods to begin a rigorous effort to identify these key features. We firstly defined an echo chamber as a network of users with highly homogeneous viewpoints and then employed an echo chamber identification algorithm to identify users within these networks. Given prior research that identified a notable influence of personality traits on the formation of echo chambers, our study innovatively integrated the five personality dimensions of the Big Five model with an additional 23 common user features, employing these as the feature variables in our analysis. Subsequently, we trained and compared the performance of three leading machine learning algorithms—LightGBM (Light Gradient Boosting Machine), XGBoost (Extreme Gradient Boosting), and CatBoost (Categorical Boosting)—in identifying echo chamber users on Weibo and Twitter (X). After selecting the best-performing model, we combined it with the SHAP (SHapley Additive exPlanations) method and successfully identified ten risk features. The results show that the CatBoost algorithm excels in accurately identifying echo chamber users. Key risk features driving users into echo chambers include user interactions with target posts, such as retweeting, commenting, and comments' sentiment and stance polarity (positive or negative). In conclusion, this study found that certain specific user behavior patterns and personality traits promote users joining echo chambers. These findings provide valuable insights into mitigating the formation and impact of echo chambers in online discussions.