WebSci’23 News Bulletin 4
15th ACM Web Science Conference: Inequalities in the Face of Concurrent Crises
30 April – 1 May 2023
Austin, Texas, USA (and online)
With less than two weeks to go until the start of this year’s Web Science Conference, here comes our third news bulletin, introducing the paper sessions and their contents happening throughout the second conference day.
Additionally, we still have some availability of free online tickets for students and early career scholars. To apply for a free ticket, please complete the online application at the link below as soon as possible but not later than Friday 21, 2023. Notifications will be sent out by Monday 24, 2023.
Link to the application form: https://forms.gle/8SvUsxB4nrGc7YEb6
In case you have not had a look at our full program yet, you may find it on our website (https://websci23.webscience.org/program/). While you are there, you may also find more details on the publications belonging to the different paper sessions (https://websci23.webscience.org/paper-sessions/).
If you should have missed it, our previous news bulletin provided an overview of the different events happening throughout the two conference days, as well as a deep dive into the paper sessions of the first conference day. You may find this one and all other news bulletins in the News section on our website.
In case you have not yet registered for the conference and would like to attend either virtually or in-person, you may still do so by following the registration instructions provided on the website (https://websci23.webscience.org/registration/).
Still need more arguments for why this conference will be well worth you while? Keep on reading below and check out what great presentations we will have during the second day of the conference! Missing info on the presentations of day 1? Look for our News Bulletin #2, we got you covered there.
For this and everything else #WebSci23, keep an eye on the hashtag on social media, and check our website for further updates.
WebSci’23 Conference Committee
Paper Session 4 (Monday, 11:00 AM – 12:30 PM): Fairness and Bias
Following the keynote by Dhiraj Murthy, right before lunch, the paper session on Fairness and Bias is the first of the three outstanding paper sessions on the second day of this year’s conference.
Ángel Pavón Pérez et al. propose an approach based on covariance analysis to identify attributes that encapsulate sensitive information in datasets used in high-stake decision-making situations, like credit and loan approvals. Their approach promises to reduce model bias while maintaining the overall performance.
Tuðrulcan Elmas studies the common issue of data decay that many Social Media datasets suffer from. By dissecting the issue into different topical contexts and looking into potentially harmful contents with particular detail, the importance of collecting this type of data in real time to avoid the development of various biases is emphasized.
The issue of bias in knowledge graphs such as Wikidata is explored by Paramita Das and colleagues. By looking into different knowledge graph embeddings for professions sampled from around the globe, they raise awareness for the need for precise design choices when it comes to data and algorithm to avoid the propagation of biases into knowledge graphs.
Weixiang Wang and Sucheta Soundarajan study the multi-faceted topic of fairness in yet another domain; Recommendation systems. Their proposed FairLink framework allows for the specification of the appropriate fairness metric for proposing new links in a link recommendation system.
Adding to the comprehensive treatment of bias in this session, Fabian Haak and Philipp Schaer present a whole suite of resources for the evaluation of bias in online search. Their dataset of Google and Bing search queries helps to understand the ways in which standard search behavior may lead to biased opinion formation.
Margherita Berte and colleagues monitor the gender gap in the Italian labor market via data from the LinkedIn Advertising Platform. Their study exposes patterns on a subnational level that also relate to what they call a digitalization gender gap.
Paper Session 5 (Monday, 1:30 PM – 3:00 PM): Harmful and Problematic Behavior
After lunch, we will hear all about the problems of Harmful and Problematic Behavior on Social Media and beyond.
Mithun Das and Animesh Mukherjee colleagues take the problem of detecting harmful content on Social Media to the visual level. Their work explores different techniques of detecting abusive memes in a multilingual and multimodal setting.
Giuseppe Russo and colleagues study the ways in which the migration of banned communities to fringe platforms effects their degree of radicalization as well as their spillover back onto the mainstream platforms they were initially banned from. Their exploration of individual- and social-level factors builds the foundation for evidence-based moderation policies going forward.
Maricarmen Arenas and colleagues take a look at the networks of sex traffickers promoting OnlyFans accounts on Twitter. Their Multi-Level Clustering method allows for the detection of networks based on features beyond the mere textual content of their Tweets.
Joseph Kwarteng and colleagues not only shed light on the phenomenon of “Misogynoir”, a very specific type of hate at the intersection of racism and sexism that is inherently difficult to detect on platforms like Twitter for its subjectivity and intersectionality, but also provide a systematic investigation of the influence of annotators’ demographics on their annotation behavior. Their work thereby highlights the relevance of annotators’ perspectives and content comprehension when it comes to tasks like labelling hate speech.
Xinyu Wang and colleagues study the racist narratives and conspiracy theories targeted at Asians in the context of the Covid-19 pandemic. Their work with data from Twitter explores how these narratives and conspiracy theories are deeply rooted in historical stereotypes, uncovering insights for improved anti-racist efforts going forward.
Hina Qayyum and colleagues cover yet another type of harmful and problematic behavior on Twitter, focusing on the users that are responsible for the largest share of toxic content that proliferates on the platform, characterizing their thematic contents and posting behaviors.
Paper Session 6 (Monday, 3:15 PM – 4:45 PM): Misinformation and Misperceptions
The last paper session of the conference – before we prepare the center stage for David Rand’s keynote, our “The Future of Web Science Panel”, and the Best Paper and WST Test of Time Awardees – is focused on the topic of Misinformation and Misperceptions.
Francesco Pieri and colleagues shed light on the spread of propaganda and misinformation on Facebook and Twitter during the Russian invasion of Ukraine. Their various analyses touch upon questions of content amplification and moderation.
Mohamad Hoseini and colleagues study the global spread of the QAnon conspiracy theory on Telegram, offering a global overview of popular themes and different linguistic communities.
Focusing on countermeasures to the problem of online misinformation, Yingchen Ma and colleagues take a closer look at the phenomenon of social correction. Their work studies the dynamics around the correction of misinformation through the intervention of other users on Twitter.
Akram Sadat Hosseini and Steffen Staab look behind the scenes of online misinformation by exploring the emotional framings of different types of claims. Taking this approach yet another step further, they also evaluate the emotional responses and sharing behavior of users seeing these claims.
Jinkyung Park and colleagues study the problems of automatic misinformation detection methods that use source level labels. Their work shows how a focus on the article level helps to conduct fairness audits and improve the detection of misinformation.
Satrio Yudhoatmojo and colleagues look at two different web communities – Reddit and 4chan – and explore how they interact with scientific work shared as e-prints. Their work shows how scientific knowledge might be misinterpreted through dissemination and discussion in these channels.