The Fifth FEVER Workshop

I will start the talk with our recent work on collecting and analyzing data from WhatsApp. I’ll try to summarize work on how WhatsApp is used by political parties in India to spread misinformation and hate speech. Next, I will delve deep into developing solutions for content moderation on WhatsApp, which is non trivial due to the end to end encrypted nature of WhatsApp. I will present two solutions: One based on an on-device setup to content moderation, before the content is encrypted, and the second based on a crowdsourced, bottom up model for fact checking. I will end with a discussion on potential future research directions in this space.

In recent years, researchers and practitioners at companies have paid significant attention to mitigating harms that can be created in the online world, such as misinformation, hate speech and other integrity violations. The primary reason we want to limit those harms is that they violate human values that we hold dear as individuals and as societies. This begs a set of broader questions: what are these human values that we should be incorporating into our online recommender systems and how can we also shift some of our attention to ensure that the recommendation systems provide benefits, and not only focus on their possible harms. In the first part of this talk, I will share some observations from a discussion on the topic of human values that included experts from the fields of AI, HCI, psychology, policy, law and journalism. In the second half of the talk I’ll describe how addressing human values also raises new challenges at the intersection of multiple fields: natural language processing, information retrieval, database and knowledge graph management.

Online social media platforms have brought numerous positive changes, including access to vast amounts of news and information. Yet, those very opportunities have created new challenges—our information ecosystem is now rife with problematic content, ranging from misinformation, conspiracy theories, to hateful and incendiary propaganda. In this talk, I will focus on one aspect of problematic online information: conspiracy theories.Leveraging data spanning millions of conspiratorial posts on Reddit, 4chan, and 8chan, I will present scalable methods to unravel who participates in online conspiratorial discussions, what causes users to join conspiratorial communities and then potentially abandon them. I will close by previewing important new opportunities to counter online misinformation, including conducting social audits to defend against algorithmically generated misinformation and designing socio-technical interventions to promote meaningful credibility assessment of information.

There are thousands of languages in the world, and they are fully understood only in their cultural context. What may be understood as an obvious fact in English may not be so obvious in another language, and social biases exacerbate this problem of miscommunication that can occur across languages. I will begin this talk with how we work on building datasets for low- and medium-resource languages. As part of that topic, I will present on-going work on annotation with language learners. I will then present research on analyzing and mitigating bias in several different languages. In the last part of the talk, I will discuss hate speech and how that can vary across languages.

The task of professionally debunking disinformation narratives is not trivial and the significant increase in information shared online has added a huge strain in the work of journalists and fact-checkers. In particular, with the COVID-19 pandemic, the WHO has coined the term "infomedic", describing a scenario where the high amount of information (including disinformation) during the pandemic outbreak, resulted in confusion and mistrust in reliable authorities. Disinformation is also known to quickly spread, highlighting the need for timely debunks and mitigation actions. In this talk, I will present Natural Language Processing (NLP) approaches for supporting the work of professional fact-checking. Drawing from the expertise of my research group on working closely with journalists and fact-checkers, I will discuss the main challenges in developing such tools bridging the gap between research and real-world applications. Finally, the importance of fairness and explainability in models' decisions will also be discussed.