https://dbsec2019.cse.sc.edu/Keynote.html Abstract: As social media permeates our daily life, there has been a sharp rise in the misuse of social media affecting our society in large. Specifically, harassment and radicalization have become two major problems on social media platforms with significant implications on the well-being of individuals as well as communities. A 2017 Pew Research survey on online harassment found that 66% of adult Internet users have observed online harassment and 41% have personally experienced it. Nearly 18% of Americans have faced severe forms of harassment online such as physical threats, harassment over a sustained period, sexual harassment or stalking. Moreover, malicious organizations (e.g., terrorist groups, white nationalists not classified legally as terrorists but as a group with extreme ideology) have been using social media for sharing their propaganda and misinformation to persuade individuals and eventually recruit them to propagate their ideology. These communications related to harassment and radicalization are complex concerning their language and contextual characteristics, making recognition of such narratives challenging for researchers as well as social media companies. As most of the existing approaches fail to capture fundamental nuances in the language of these communications, two prominent challenges have emerged: ambiguity and sparsity. Sole data level bottom-up analysis has been unsuccessful in revealing the actual meaning of the content. Considering the significant sensitivity of these problems and its implications at individual and community levels, a potential solution requires reliable algorithms for modeling such communications. Our approach to understanding communications between source and target requires deciphering the unique language, semantic and contextual characteristics, including sentiment, emotion, and intention. This context-aware and knowledge-enhanced computational approach to the analysis of these narratives breaks down this long-running and complex process into contextual building blocks that acknowledge inherent ambiguity and sparsity. Based on prior empirical and qualitative research in social sciences, particularly cognitive psychology, and political science, we model this process using a combination of contextual dimensions -- e.g., for Islamist radicalization: religion, ideology, and hate -- each elucidating a degree of radicalization and highlighting independent features to render them computationally accessible.