Suchergebnisse

Hate speech is a phenomenon studied in numerous disciplines by many researchers. This research has produced a variety of findings, e.g., with regard to the prevalence of hate, common targets or differences between platforms or countries. However, previous research also comes with conceptual and methodological challenges, e.g., definitions or operationalizations of hate speech in empirical studies. The present chapter focuses on the issue of equivalence in previous hate speech research - a well-known problem of comparative research in general. To compare research findings relating to hate speech across different contexts scholars need to consider the equivalence with respect to definitions, methods, measurements, procedures, and also the sampling communication content. We provide an overview about potential pitfalls and biases that can be due to a lack of equivalence and point to strategies on how to address them.

This chapter discusses the origins, development, and characteristics the hate network built around the President of Brazil Jair Bolsonaro. How is it structured? How does it work? What factors have allowed it to exist? How does it affect the health of Brazilian democracy? It argues that Bolsonaro's followers took advantage from an institutional crisis taking place in the late 2010s. At that time, the legacy media provided a massive coverage associating PT (Workers' Party), which was ahead of the presidency, to corruption, thus fostering a hate campaign against it. In that scenario, Bolsonaro emerged as the leader of the far-right opposition to PT. His followers firstly built a network on Facebook, using fake profiles and mixing hate speech with humor. Coordinated unofficially by several cabinet staff members, this network articulates the official profiles of the Bolsonaro family to followers and sympathizers and a vast array of anonymous supporting pages. This structure allowed Bolsonaro's activists to blur the boundaries between official and spurious discourses, and powered a series of flaming wars against political adversaries and reporters perceived as hostile to Bolsonaro's interests.

Incivility is a concept with a wide scope of interpretations, ranging from impoliteness to aggressive and extremist speech. The definition of uncivil speech is highly context-sensitive, and this contextual sensitivity should be considered in future research. In this chapter, I argue that choosing to omit context from incivility research may result in the diffusion of authoritarian norms in online content regulation and negatively influence freedom of speech in different sociopolitical settings. I suggest considering four layers of context in incivility research: (1) sociocultural context (the macro level), (2) sociopolitical context (the macro level), (3) organizational context (the meso level), and (4) situational context (the micro level). I elaborate on each level's role in defining and regulating uncivil speech, and I conclude by suggesting paths for future research.

Effective and rights-respecting hate speech interventions should encourage cohesion more than hate and inclusion more than division. Importantly, they should also guarantee the right to be and to express. Unfortunately, most countries, including Nigeria, lack effective hate speech interventions. This chapter considers specific ways to make hate speech interventions in Nigeria more effective while guaranteeing the protection of the right to freedom of expression, especially in the digital age. Thus, this chapter considers international human rights law and ideas on hate speech interventions, various hate speech interventions in Nigeria, and Nigeria's obligation to comply with international human rights instruments to achieve better results. It concludes that one of the best ways to ensure these interventions' effectiveness is combining alternative methods, such as strategic training, education, public awareness, and a multistakeholder approach.

Building on the growing concerns around hate speech and harmful content on social media, this chapter analyzes the processes by which content is moderated on leading social media platforms. The outsourcing practices of platform operators or social media companies to acquire content moderation services from third-party companies have been acknowledged in the public discourse. Details regarding these outsourcing relationships and power mechanisms remain obfuscated, however. Using empirical data from India, this chapter presents a global value chain perspective on the mechanisms by which US-based social media monopolies source content moderation services from Indian information technology business process outsourcing (IT BPO) supplier companies. The agreements established between the two parties direct the content moderation labor process through which Indian workers' labor power is transformed into productive labor.

The last decade saw the rise of research in the area of hate speech and abusive language detection. A lot of research has been conducted, with further datasets being introduced and new models put forward. However, contrastive studies of the annotation of different datasets also revealed that some problematic issues remain. Theoretically ambiguous and misleading definitions between different studies make it more difficult to evaluate model reproducibility and generalizability and require additional steps for dataset standardization. To overcome these challenges, the field needs a common understanding of concepts and problems such that standard datasets and different compatible approaches can be developed, avoiding inefficient and redundant research. This article attempts to identify persistent challenges and develop guidelines to help future annotation tasks. Some of the challenges and guidelines identified and discussed in the article relate to concept subjectivity, focus on overt hate speech, dataset integrity and lack of ethical considerations.

In research on online comments on social media platforms, different terms are widely used to describe comments that are hateful or disrespectful and thereby poison a discussion. This chapter takes a theoretical perspective on the term toxicity and related research in the field of computer science. More specifically, it explains the usage of the term and why its exact interpretation depends on the platform in question. Further, the article discusses the advantages of toxicity over other terms and provides an overview of the available toxic comment datasets. Finally, it introduces the concept of engaging comments as the counterpart of toxic comments, leading to a task that is complementary to the prevention and removal of toxic comments: the fostering and highlighting of engaging comments.

The advent of social media has increased digital content - and, with it, hate speech. Advancements in machine learning help detect online hate speech at scale, but scale is only one part of the problem related to moderating it. Machines do not decide what comprises hate speech, which is part of a societal norm. Power relations establish such norms and, thus, determine who can say what comprises hate speech. Without considering this data-generation process, a fair automated hate speech detection system cannot be built. This chapter first examines the relationship between power, hate speech, and machine learning. Then, it examines how the intersectional lens - focusing on power dynamics between and within social groups - helps identify bias in the data sets used to build automated hate speech detection systems.

To quickly identify hate speech online, communication research offers a useful tool in the form of automatic content analysis. However, the combined methods of standardized manual content analysis and supervised text classification demand different quality criteria. This chapter shows that a more substantial examination of validity is necessary since models often learn on spurious correlations or biases, and researchers run the risk of drawing wrong inferences. To investigate the overlap of theoretical concepts with technological operationalization, explainability methods are evaluated to explain what a model has learned. These methods proved to be of limited use in testing the validity of a model when the generated explanations aim at sense-making rather than faithfulness to the model. The chapter ends with recommendations for further interdisciplinary development of automatic content analysis.

This chapter discusses the legal aspects of researchers' access to social media data, focusing in particular on recent developments in European law. We see law as playing both an enabling and a restrictive role in facilitating platform data access. Identifying a number of shortcomings in current legislation, we argue for the creation of a sound legal framework for scholarly data research. The new Digital Services Act makes some promising first steps towards regulating programmatic data access through APIs, but many obstacles and ambiguities remain. Furthermore, a clear vision on the legal status of public interest scraping projects is still lacking. In the teeth of private ordering by global platform companies, as new gatekeepers in academic research, ensuring fair and rights-sensitive data access must be a priority for the (European) legislator.

Working with social media data is a hermeneutic procedure systematically guided by doubts about the meaning of data at all stages of the research process, from data collection and preparation to data analysis and publication. A short walk through the automated data collection workflow, as it is implemented in the open-source software Facepager, highlights some of the epistemic peculiarities of the process. The paper encourages researchers to deal with technical details, errors, and restrictions in order to gain a deeper understanding of the organizing principles of the web. Technical limitations and hurdles should not solely be considered as problems to be solved, but also as indicators of social processes on online platforms. Scraping social media data touches on key aspects of platformization and, therefore, is not merely a data collection method, but also a means of examining the online world through a data hermeneutical lens.

As long as we have attempted to sanction untoward speech, others have devised strategies for expressing themselves while dodging such sanctions. In this intervention, I review the arms race between technological filters designed to curb hate speech, and evasive language practices designed to avoid detection by these filters. I argue that, following important advances in the detection of relatively overt uses of hate speech, further advances will need to address hate speech that relies on culturally or situationally available context knowledge and linguistic ambiguities to convey its intended offenses. Resolving such forms of hate speech not only poses increasingly unreasonable demands on available data and technologies, but does so for limited, uncertain gains, as many evasive uses of language effectively defy unique valid classification.

Extreme speech is a critical conceptual framework that aims to uncover vitriolic online cultures through comparative and ethnographic excavations of digital practices. It is not one more new definition or a term replaceable with extremist speech. Rather, it is a conceptual framework developed to foreground historical awareness, critical deconstruction of existing categories, and a grounded understanding of evolving practices in online communities, in ways to holistically analyze the contours and consequences of contemporary digital hate cultures. This framework suggests that the close contextualization of proximate contexts - of media affordances in use or situated speech cultures - should accompany deep contextualization, which accounts for grave historical continuities and technopolitical formations unfolding on a planetary scale. Through such elaborate forays into everyday practices and deeper histories, extreme speech theory proposes to nuance normative and regulatory efforts to classify and isolate hate speech and disinformation.

Current definitions of hate speech are inadequate as the basis for monitoring hate speech targeted at groups. First, they do not capture escalating group-targeted negative speech which can be a precursor to more extreme forms of hate speech such as dehumanization, demonization, and incitement to violence. While not hate speech, such negative speech is an early warning that could be helpful for a hate speech monitoring system to track, as responses and interventions, especially to the offline harms of hate speech, can take time to operationalize. Second, current definitions of hate speech do not capture hateful rhetoric aimed at groups not traditionally included in hate speech definitions (those without immutable qualities), such as groups targeted for hate based on profession-based identity like journalists. This chapter presents some suggestions for addressing these issues, including a hate speech intensity scale.

Hate speech has been identified as a pressing problem in society, and several automated approaches have been designed to detect and prevent it. This chapter reflects on the operationalizations, transformations, and reductions required by the datafication of hate to build such an automated system. The observations are based on an action research setting during a hate speech monitoring project conducted in a multi-organizational collaboration during the Finnish municipal elections in 2017. The project developed an adequately well-working algorithmic solution using supervised machine learning. However, the automated approach requires heavy simplification, such as using rudimentary scales for classifying hate speech and relying on word-based approaches, while in reality hate speech is a nuanced linguistic and social phenomenon with various tones and forms. The chapter concludes by suggesting some practical implications for developing hate speech recognition systems.