In the past decade, governments, media companies, and communities have intensified efforts to combat hate speech and aggression online through policies, trainings, and educational programming. Nevertheless, they are machine learning (ML) tools, algorithms that learn to identify hate speech, which remain the central component of media companies’ strategies for reducing online harms. Algorithms have become more and more accurate in targeting hate speech, however their development has primarily focused on improving ML capacity to identify linguistic and textual features flagged as hate speech. Further, variation in linguistic expression and the lack of large datasets for training algorithms to search for these expressions limit the efficacy of ML tools. Bringing together experts in geo-spatial mapping, digital media and linguistic analysis, and computer scientists, project participants will contribute to building a software that helps researchers and companies to understand the interactions and user engagements surrounding instances of hate speech. Using a custom designed network analytic software, we build a connectivity-defined architecture for mapping online Twitter communities in order to identify spatial, temporal, discursive and engagement patterns in the circulation of hate speech within select Twitter communities. Analyzing how users within and outside of particular Twitter communities engage with an instance of hate speech over time (through types of contact, conflict, reaction and engagement)—data collected on these specific connectivities surrounding hate speech have the potential to inform and improve ML hate speech detection.