arXiv subjects

Mona Elswah

Publications and source records attributed to Mona Elswah.

3 recordsLinked to original sources

arXiv2025

Think Outside the Data: Colonial Biases and Systemic Issues in Automated Moderation Pipelines for Low-Resource Languages

Most social media users come from the Global South, where harmful content usually appears in local languages. Yet, AI-driven moderation systems struggle with low-resource languages spoken in these regions. Through semi-structured interviews with 22 AI experts working on harmful content detection in four low-resource languages: Tamil (South Asia), Swahili (East Africa), Maghrebi Arabic (North Africa), and Quechua (South America)--we examine systemic issues in building automated moderation tools for these languages. Our findings reveal that beyond data scarcity, socio-political factors such as tech companies' monopoly on user data and lack of investment in moderation for low-profit Global South markets exacerbate historic inequities. Even if more data were available, the English-centric and data-intensive design of language models and preprocessing techniques overlooks the need to design for morphologically complex, linguistically diverse, and code-mixed languages. We argue these limitations are not just technical gaps caused by "data scarcity" but reflect structural inequities, rooted in colonial suppression of non-Western languages. We discuss multi-stakeholder approaches to strengthen local research capacity, democratize data access, and support language-aware solutions to improve automated moderation for low-resource languages.

cs.CL↗

arXiv2025

Bridging Boundaries: How to Foster Effective Research Collaborations Across Affiliations in the Field of Trust and Safety

As the field of Trust and Safety in digital spaces continues to grow, it has become increasingly necessary - but also increasingly complex - to collaborate on research across the academic, industry, governmental and non-governmental sectors. This paper examines how cross-affiliation research partnerships can be structured to overcome misaligned incentives, timelines and constraints while delivering on the unique strengths of each stakeholder. Drawing on our own experience of cross-sector collaboration, we define the main types of affiliation and highlight the common differences in research priorities, operational pressures and evaluation metrics across sectors. We then propose a practical, step-by-step framework for initiating and managing effective collaborations, including strategies for building trust, aligning goals, and distributing roles. We emphasize the critical yet often invisible work of articulation and argue that cross-sector partnerships are essential for developing more ethical, equitable and impactful research in trust and safety. Ultimately, we advocate collaborative models that prioritize inclusivity, transparency and real-world relevance in order to meet the interdisciplinary demands of this emerging field.

cs.CY↗

arXiv2021

Look who's watching: platform labels and user engagement on state-backed media outlets

Recently, social media platforms have introduced several measures to counter misleading information. Among these measures are state media labels which help users identify and evaluate the credibility of state-backed news. YouTube was the first platform to introduce labels that provide information about state-backed news channels. While previous work has examined the efficiency of information labels in controlled lab settings, few studies have examined how state media labels affect user perceptions of content from state-backed outlets. This paper proposes new methodological and theoretical approaches to investigate the effect of state media labels on user engagement with content. Drawing on a content analysis of 8,071 YouTube comments posted before and after the labelling of five state-funded channels (Al Jazeera English, CGTN, RT, TRT World, and Voice of America), this paper analyses the effect state media labels had on user engagement with state-backed media content. We found the labels had no impact on the amount of likes videos received before and after the policy introduction, except for RT which received less likes after it was labelled. However, for RT, comments left by users were associated with 30 percent decrease in the likelihood of observing a critical comment following the policy implementation, and a 70 percent decrease in likelihood of observing a critical comment about RT as a media source. While other state-funded broadcasters, like Al Jazeera English and VOA News, received fewer critical comments after YouTube introduced its policy; this relationship was associated with how political the video was, rather than the policy change. Our study contributes to the ongoing discussion on the efficacy of platform governance in relation to state-backed media, showing that audience preferences impact the effectiveness of labels.

cs.SI↗