The dissemination of online hate speech can have serious negative consequences for individuals, online communities, and entire societies. This and the large volume of hateful online content prompted both practitioners’, e.g., in content moderation or law enforcement and researchers’ interest in machine learning models to automatically classify instances of hate speech. Whereas most scientific works address hate speech classification as a binary task, practice often requires a differentiation into sub-types, e.g., according to target, severity, or legality, which may overlap for individual content. Hence, researchers created datasets and machine learning models that approach hate speech classification in textual data as a multi-label problem. This work presents the first systematic and comprehensive survey of scientific literature on this emerging research landscape in English (N = 46). We contribute with a concise overview of 28 datasets suited for training multi-label classification models, revealing significant heterogeneity regarding label-set, size, meta-concept, annotation process, and inter-annotator agreement. Our analysis of 24 publications proposing suitable classification models further establishes inconsistency in evaluation and a preference for architectures based on Bidirectional Encoder Representation from Transformers (BERT) and Recurrent Neural Networks (RNNs). We identify imbalanced training data, reliance on crowdsourcing platforms, small and sparse datasets, and missing methodological alignment as critical open issues and formulate 12 recommendations for research. Whereas transparency, data source diversity, conceptual rigor, label-set consolidation, inter-annotator agreement, and model validation on existing datasets warrant more attention, multi-label tasks related to criminal relevance, enhancing robustness through ensembling, and exploring performance gains through large foundation models represent promising avenues for future work.

Youth are particularly likely to encounter hateful internet content, which can severely impact their well-being.While most social media provide reporting mechanisms, in several countries, severe hateful content can alternatively be reported to law enforcement or dedicated reporting centers.However, in Germany, many youth never resort to reporting.While research in human-computer interaction has investigated adults' views on platform-based reporting, youth perspectives and platform-independent alternatives have received little attention.By involving a diverse group of 47 German adolescents and young adults in eight focus group interviews, we investigate how youth-sensitive reporting systems for hateful content can be designed.We explore German youth's reporting barriers, fnding that on platforms, they feel particularly discouraged by defcient rule enforcement and feedback, while platform-independent alternatives are rather unknown and perceived as time-consuming and disruptive.We further elicit their requirements for platformindependent reporting tools and contribute with heuristics for designing youth-sensitive and inclusive reporting systems.

In Germany and other countries, specialized non-profit reporting centers combat online hate speech by submitting criminal content to law enforcement agencies, forwarding deletion requests to social media platforms, and providing counseling to victims, thus contributing to the governance mechanism of content moderation as intermediaries between victims and various organizations. Whereas research in computer-supported cooperative work has extensively explored collaboration of and automation for content moderators, there are no works that focus on reporting centers. Based on expert interviews with their staff (N=15), this study finds that most German centers share a collaborative workflow, of which multiple tasks are heavily dependent on inter-organizational exchange. However, there are differences in their implementation of monitoring, content assessment, automation technology adoption, and external collaborators. As the centers are faced with diverse challenges, such as borderline case assessment, psychological burdens, limited visibility, conflicting goals with other actors, and manual repetitive work, our study contributes with nine implications for designing and researching supportive technologies. They provide suggestions for improving hate speech gathering and reporting, researching hate speech prioritization and assessment algorithms, and designing case processing systems. Beyond that, we outline directions for research on inter-organizational collaboration.

Abstract The negotiation of stakeholder values as a collaborative process throughout technology development has been studied extensively within the fields of Computer Supported Cooperative Work and Human-Computer Interaction. Despite their increasing significance for cybersecurity incident response, there is a gap in research on values of importance to the design of open-source intelligence (OSINT) technologies for this purpose. In this paper, we investigate which values and value conflicts emerge due to the application and development of machine learning (ML) based OSINT technologies to assist cyber security incident response operators. For this purpose, we employ a triangulation of methods, consisting of a systematic survey of the technical literature on the development of OSINT artefacts for cybersecurity (N = 73) and an empirical value sensitive design case study, comprising semi-structured interviews with stakeholders (N = 9) as well as a focus group (N = 7) with developers. Based on our results, we identify implications relevant to the research on and design of OSINT artefacts for cybersecurity incident response.