Introduction
In recent years, the rapid growth of social media platforms across the globe has changed the way people communicate. Online social networking and micro-blogging sites like Twitter, Facebook attract internet users more than any other type of site （Watanabe et al., 2018). However, these platforms also pose a number of challenges, especially when it comes to the management of hate speech. Hate speech is an artistic term that refers to a specific expression of hatred against a particular group in a specific context (Howard, 2019). The spread of hate speech not only affects the mental health of individuals, but can also incite violence and seriously destroy social stability. At present, the problem of hate speech is prevalent on social platforms, especially on global platforms such as Facebook and Twitter.
This article will deeply analysis the responsibilities and challenges faced by social media platforms, when dealing with hate speech. And analyze the case studies of Facebook and Twitter, discuss their strategies and effects in dealing with hate speech.
Challenges for platforms to regulate hate speech
The content creation and sharing mechanisms of social media platforms have become a breeding ground for hate speech. In both legally and ethically perspectives, platforms should take more responsibility for the spread of hate speech. User-generated content is one of the core characteristics of social platforms, but this model that allows any user to post, share, and disseminate information can easily lead to the spread of harmful speech, especially on sensitive social issues. Initially, platforms believed that users should manage their own speech. As social and political pressures increase, and platforms become more influential in society, hate speech is more clearly defined, platforms are beginning to acknowledge the need to take more responsibility for these issues (Dubois & Reepschlager, 2024).
Today, algorithms are commonly used by platforms to manage hate speech. But on the one hand, computers must be evaluated according to rules, including laws, culture, customs, business, and so on, dividing a large number of things into acceptable and unacceptable have far exceeded the capabilities of AI (Roberts, 2019). AI systems can censor large amounts of content in a short time, quickly spotting hate speech in keywords, images, or videos. However, this approach has its obvious limitations, especially when facing complex cultural and linguistic differences, AI often misjudges. For example, certain speech may be considered harmless in one cultural context, but in another language, it is hate speech. Watanabe et al (2018) found that features such as the semantic features they analyze need to be used combine with other features to be meaningful, at the same time, offensive and hateful content is difficult to distinguish. On the other hand, the platform’s AI system is probably be “control”. Since the funds and methods of building and optimizing AI systems are all aimed at serving the existing dominant interests, it can be said that “Artificial intelligence is a registry of power”(Crawford, 2021). The operation of the AI system is actually the process of processing the data, so the data is the embodiment of resources and power. Behind the algorithm, there are national interests, platform interests and other interests driven. At this time, AI is not only a technical tool, but also carries the power to set social rules and moral standards. Therefore, platforms not only has the responsibility to maintain fairness, but also need to maintain transparency, and protect the data and privacy of users.
To compensate for the lack of AI, social platforms also rely on human teams to further censor sensitive content. These human censors often process on content that the AI fails to clean up, to ensure that it complies with the platform’s policies. This work is heavy and complex, so human teams must have a high level of professional judgment. Human censor is easily limited by human subjective factors, but in general ,we should follow the point of Howard (2019), he proves that speakers have five duties to avoid hate speech as follow:
- Not to Threaten;
- Not to Harass;
- Not to Offend;
- Not to Defame;
- Not to Incite Wrongdoing.
Furthermore, It is also difficult for platforms to balance freedom of speech with public safety. In many countries and regions, freedom of expression is considered a fundamental human right. Curbing hate speech through legal procedures is a fight against the right to freedom of expression in national constitutions (Sinpeng et al., 2021). Platforms that heavily censor hate speech can infringe on users’ freedom of expression. Excessive censorship can lead to restrict normal communication of speech. Social platforms need to find the balance between protecting freedom of expression and ensuring that their content does not cause harm to others.
Facebook vs. Twitter Hate Speech Management
Facebook introduced its Terms of Use in 2004 and replaced them with a Statement of Rights and Responsibilities in 2009, similarly, the Twitter Rules were introduced in 2006 and the Abuse Guidelines were added in 2013 (Dubois & Reepschlager, 2024). Facebook and Twitter have adopted a self-regulatory approach in response to hate speech, They developed content management rules and set up a team to regulate and operate them. But the commercial interests of the platform may conflict with the regulatory responsibilities. For example, platforms want to increase user engagement by recommending extreme content, which may be at odds with their content management policies.
Facebook has been repeatedly criticized by the public and the government in the face of hate speech. In the period of covid-19 pandemic, there has been a surge in discriminatory and hate speech against China on Facebook. Many social media users have linked the outbreak to China, leading to hate attacks on Asian groups, especially in posts and groups discussing the corona-virus. Although Facebook explicitly prohibits hate speech and discriminatory speech in its policies, the platform failed to clean up these hate speech in early 2020 timely. Thus, some posts revile China, calling for punishment or violence against the “source of the virus”. These comments were not detected and deleted by Facebook timely, resulting in a large number of users being harassed and discriminated against. Some content that is considered hate speech does not fit Facebook’s definition of hate speech, therefore it was not processed (Sinpeng et al., 2021). Facebook’s failure to pay attention to non-English languages is allowing hate speech to flourish (theconversation.com) shows that while Facebook has broadened its definition of hate speech, it often refuses to respond to report about hate speech, and some hate speech cannot be identified due to language.
Even today, in 2025, when I search for covid-19 on Facebook, I observed that hate speech about China, “not to defamation”, and it reads that “97% percent of the members of the French Academy of Medicine believe that the covid-19 virus did not arise naturally, but came from a laboratory in Wuhan, China.”(the post) The user made up something that didn’t exist, but his account was still available and the content he posted was not deleted. After I reported him, I didn’t see any changes to his page. Facebook’s page admins discovered that the hate speech content they flagged was not dealt with immediately. Facebook needs to provide more mandatory training for page admins on identifying and managing hate speech (Sinpeng et al., 2021).
After receiving feedback from users and organizations, Facebook has strengthen its detection and removal measures against covid-related hate speech. Facebook has strengthened its collaboration with the Asian community and increased transparency, publishing regular audit reports to show the results of anti hate speech to the public. The platform’s increased transparency can give users more security and make regulation more powerful.
Facebook’s response has exposed the platform’s inadequacies in responding to public emergencies. In particular, language differences, cultural background differences, and the legal environment of different countries around the world make it extremely challenging for Facebook to respond quickly. It shows that social platforms need to adapt the difference on a global scale. It is important to cooperate between platforms, and platforms need to cooperate with governance. But when it comes to cross-border cooperation, the issues become more complex and difficult to operate. For example, incitement to racial or religious hatred is a criminal offence in developed democracies such as the United Kingdom, Australia, Denmark, France, Germany and so on (Waldron, 2012). But this situation is unconstitutional in the United States (Howard, 2019).
Similar to Facebook, Twitter has faced a complex set of issues in its response to hate speech, especially during the Trump presidency, when it became a major platform for political speech. In particular, Trump’s speech about immigrant, race, gender and religious groups have often been controversial. Some speech have not only stirred up the emotions of extreme supporters, but also led to a surge in hate speech against opponents. As one of the world’s most well-known social media platforms, how Twitter responds to these issues become the focus of public attention. In May 2020, Trump posted a tweet, “When the looting starts, the shooting starts.” The tweet stirred up hostility towards black protesters, increased the tension in American society. Twitter flagged the tweet as violating the platform’s “incitement to violence” policy, but did not immediately remove the content. The delay sparked strong public criticism. Until January 2021, Twitter permanently banned Trump’s account.
In addition, Twitter’s response highlights an important question: how to balance freedom of expression with public safety. At the same time, it shows the regulatory dilemma of platforms in the face of globally well-known public figures. Although Twitter eventually showed strong regulatory determination by banning Trump’s account, the delay in banning Trump’s account for a long time is unfair to other ordinary users, the platform does not adopt uniform standards and means for Trump and the public. While Twitter needs to protect free speech, it should be more strict when facing public figures’ speeches, as public figures will have more attention and greater social impact.
Facebook and Twitter face similar challenges and responsibilities when it comes to strategies for dealing with hate speech — how to balance free speech with user safety, how to deal with content that incites hatred, and how to better regulate it globally. Changing policy structures, posting habits, and multiple languages complicate data tracking, research shows that Facebook provides users with tracking policy changes while Twitter does not (Dubois & Reepschlager, 2024). Both platforms use a combination of automated and human approaches to reviewing user-generated content. From the case of Facebook, we can see the platform’s efforts to improve in terms of transparency, regulation, and cooperation. As a more open real-time platform, Twitter is facing huge pressure to moderate, and should optimize its content moderation mechanism and policies.
future prospects
To improve the platform’s management of hate speech, platforms can:
- Optimize AI’s ability to identify hate speech, especially across cultures and languages.
- Cooperate with governments, organizations, and other platforms to develop globally applicable guidelines and strategies for hate speech regulation.
- Establish a clearer and more transparent content moderation policy, disclose its content moderation processes to ensure fairness and transparency.
- Strengthen user education, encourage users to reflect on the negative impact of hate speech.
- Strengthen professional education for manual review.
Conclusion
Among the hate speech regulation issues faced by social media platforms, platforms need to balance free speech with public safety. While AI technology has make content screening easier, the limitations of cultural, linguistic differences and so on still need to be supplemented by human review. Facebook and Twitter, as the world’s leading platforms, have taken regulatory measures to address hate speech, but they have also exposed issues such as platform conflicts of interest, untimely processing, limitation of cross-cultural regulation, unfair. In the future, platforms should optimize AI identification capabilities, promote global cooperation, formulate unified regulatory policies, strengthen transparency and user education, to create a safer and fairer online environment.
Reference
Crawford, K. (2021). The Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence (pp. 1-21). Yale University Press.
https://ebookcentral.proquest.com/lib/usyd/reader.action?docID=6478659&ppg=21
Dubois, E., & Reepschlager, A. (2024). How harassment and hate speech policies have changed over time: Comparing Facebook, Twitter and Reddit (2005–2020). Policy and Internet, 16(3), 523–542.
https://doi.org/10.1002/poi3.387
Howard, J. W. (2019). Free Speech and Hate Speech. Annual Review of Political Science, 22(1), 93–109.
https://doi.org/10.1146/annurev-polisci-051517-012343
Misha, K. (2021, July 5). Facebook’s failure to pay attention to non-English languages is allowing hate speech to flourish. The Conversation.
https://theconversation.com/facebooks-failure-to-pay-attention-to-non-english-languages-is-allowing-hate-speech-to-flourish-163723
Roberts, S. T. (2019). Behind the screen : Content moderation in the shadows of social media. Yale University Press.
https://ebookcentral.proquest.com/lib/usyd/detail.action?docID=5783696.
Sinpeng, A., Martin, F. R., Gelber, K., & Shields, K. (2021). Facebook: Regulating Hate Speech in the Asia Pacific. Department of Media and Communications, The University of Sydney.
https://hdl.handle.net/2123/25116.3
Waldron, J. (2012). The harm in hate speech. Harvard University Press.
Watanabe, H., Bouazizi, M., & Ohtsuki, T. (2018). Hate Speech on Twitter: A Pragmatic Approach to Collect Hateful and Offensive Expressions and Perform Hate Speech Detection. IEEE Access, 6, 13825–13835. https://doi.org/10.1109/ACCESS.2018.2806394
Be the first to comment