User-generated content within online social networks, discussion forums, and news media provides a wide-ranging mirror of society and has become a vital data source for researchers. With the constant emergence of novel platforms and the ever-changing nature of communication modalities, user participation and discussion are continuously evolving. Each new generation of users is more likely to adapt to the newer platforms rather than join older platforms, requiring new observations and insights. User demographics and location play a particular role in platform adaptation and usage; however, offline events such as elections, political discussions, and crises lead to extensive discourse on such platforms. Such events trigger an inertial moment that brings people into a large-scale and multi...[ Read more ]

User-generated content within online social networks, discussion forums, and news media provides a wide-ranging mirror of society and has become a vital data source for researchers. With the constant emergence of novel platforms and the ever-changing nature of communication modalities, user participation and discussion are continuously evolving. Each new generation of users is more likely to adapt to the newer platforms rather than join older platforms, requiring new observations and insights. User demographics and location play a particular role in platform adaptation and usage; however, offline events such as elections, political discussions, and crises lead to extensive discourse on such platforms. Such events trigger an inertial moment that brings people into a large-scale and multi-faceted online discourse; such discourse can result in a polarized community or a collective effort for social activism. These processes eventually form an opinion among the masses and may generate societal bias toward specific issues.

There is extensive literature on social media usage and bias mitigation on user-generated content across different demographics. However, there needs to be more focus on the characterization of non-trivial and dynamic content and user behavior. For instance, there is limited literature on niche social media platforms like Instagram (popular for lifestyle photography) for social activism or more specific online communities such as WallStreetBets (famous for stock trading) on Reddit concerning offline events. In addition to the characterization of platforms and their users in dynamic settings such as for social activism, a critical question is how we can account for human bias while studying this highly subjective user-generated content. Human bias and subjectivity can lead to different interpretations of the same content. Insights from such studies remain useful for disciplines beyond Computer Science. This thesis focuses on characterizing online platforms (Instagram, Reddit, and Twitter) usage during major social events and proposing a method to account for human bias when studying such user-generated content.

This thesis contains four main contributions: 1) We study the use of the Instagram platform for social activism. We take a case study of social unrest and highlight that users circumvent the platform limitations by using screenshots and embedding their messages into various symbols. 2) We study the surge in WallStreetBets during the GameStop Short Squeeze event and explore newly subscribed users' quick adoption of community communication norms. 3) We characterize Twitter users and their content focusing on their follower change during the 2022 Russia-Ukraine crisis. 4) Finally, we provide a detailed summary of how the human interpretation of any subjective content can be subject to comprehension bias; hence, we propose a method to counter this human bias with a use case of political content.

This work presents new insights into the users' adoption of the platforms, regardless of their primary or commonly perceived goals, and highlights the larger potential of such platforms. We also highlight the behavioral features that correlate with users' follower change during the crisis. These insights can be used for downstream research such as prediction tasks, along with the generalizability and applicability of the bias detection method in real-life crowdsourcing systems.