With the rise of various social media platforms and the democratization of data, online content has become significantly more accessible to users. However, with a larger user base and lax content moderation by these platforms, the spread of harmful content has also increased, particularly during elections and crises. User interaction with harmful content often leads to psychological and emotional distress. This work focuses on investigating harmful content in connection with three events: the COVID-19 pandemic, the 2019 European Parliament election in Germany, and the 2022 Russian-Ukrainian conflict. The work collects datasets using the AMUSED framework via the official application programming interfaces (APIs) of Twitter (now X), YouTube, and Facebook to examine the spread of harmful content on these platforms in relation to the three events. This work focuses on two types of harmful content: misinformation and hate speech, primarily examining the detection, dissemination, and user interaction with this content. The added value of this work can be divided into three areas: methodological, theoretical, and practical/societal. The methodological added value includes the development of classification models using machine learning methods to detect harmful content by utilizing textual, network-based, and social features. A tuned language model (LLM) is used to identify media frames and analyze cross-platform spillover effects related to the 2022 Russian-Ukrainian conflict. Furthermore, novel approaches such as engagement scores and propagation velocity are proposed to measure user interaction and information dissemination on social media. The theoretical contributions include the application of existing theories on framing research, such as Fink's model for describing the phases of the COVID-19 pandemic, agenda-setting theory for analyzing the political agendas of various political actors, and framing theory for identifying media frames and their dissemination via social media. Societal/practical media frame contributions include the role of social media in content moderation, user awareness-raising, and the presentation of a prototype for future social media that mitigates the spread of harmful content. Ultimately, by combining theoretical, algorithmic, and practical/societal contributions, the work summarizes the findings of nine research papers and provides recommendations and guidelines for counteracting harmful content. Furthermore, the work contributes to research by proposing a novel approach to analyzing harmful content, which uses various combinations of features and can be easily applied to other events or other social media platforms. In summary, the key points are the spread of harmful content on various social media platforms and events, and the investigation of the role of machine learning models, content moderation, and user awareness in combating the spread of misinformation.
Gautam Kishore Shahi (Wed,) studied this question.