WEBSCI '24: ACM Web Science Conference

The study of opinions towards migrants is profoundly important to understanding migration as well as to politics. Previous research has contributed to understanding anti-immigrant attitudes using social media data. However, there is still a need for a better understanding of opinions towards migrants in transit. We study the case of Central American migrant caravans from 2018 to 2021 by looking at the opinions in both the US, the destination country, and Mexico, the transit country. Media highly covered these events, and an online debate about them started on social media. Our research aims to understand how migrant caravans are discussed online. We are particularly interested in how media salience and geographical variables are associated with the sentiment intensity of the opinions. We combine geolocated data from Twitter, GDELT (Global Database of Events, Language, and Tone), and Survey and Census data for the US and Mexico. We use topic modeling to find the latent topics within the online Twitter discussion, and VADER sentiment analysis to quantify tweets' sentiments to calculate the sentiment intensity score that is used as the dependent variable of our OLS regression models. For both countries, we found that similar topics were discussed with a more political discussion in the US. Our analysis of the sentiment score revealed that sentiment does not reflect stance adequately, which led us to analyze the sentiment intensity score (absolute value of sentiment). We found that, for Mexico, when the media generated a higher number of news articles about migrant caravans, the sentiment intensity was higher. For the geographical variables, we found no significant association in the US; however, for Mexico, tweets in bordering states had a lower sentiment intensity. These results shed light on the differences in the determinants of sentiment intensity in opinions between the two countries.

The presence of political misinformation and ideological echo chambers on social media platforms is concerning given the important role that these sites play in the public’s exposure to news and current events. Algorithmic systems employed on these platforms are presumed to play a role in these phenomena, but little is known about their mechanisms and effects. In this work, we conduct an algorithmic audit of Twitter’s Who-To-Follow friend recommendation system, the first empirical audit that investigates the impact of this algorithm in-situ. We create automated Twitter accounts that initially follow left and right affiliated U.S. politicians during the 2022 U.S. midterm elections and then grow their information networks using the platform’s recommender system. We pair the experiment with an observational study of Twitter users who already follow the same politicians. Broadly, we find that while following the recommendation algorithm leads accounts into dense and reciprocal neighborhoods that structurally resemble echo chambers, the recommender also results in less political homogeneity of a user’s network compared to accounts growing their networks through social endorsement. Furthermore, accounts that exclusively followed users recommended by the algorithm had fewer opportunities to encounter content centered on false or misleading election narratives compared to choosing friends based on social endorsement.

Global events like wars and pandemics can intensify online discussions, fostering information sharing and connection among individuals. However, the divisive nature of such events may lead to polarization within online communities, shaping the dynamics of online interactions. Our study delves into the conversations within the largest Italian and French Reddit communities, specifically examining how the Russian invasion of Ukraine affected online interactions. We use a dataset with over 3 million posts (i.e., comments and submissions) to (1) describe the patterns of moderation activity and (2) characterize war-related discussions in the subreddits. We found changes in moderators’ behavior, who became more active during the first month of the war. Moreover, we identified a connection between the daily sentiment of comments and the prevalence of war-related discussions. These discussions were not only more negative and toxic compared to non-war-related ones but also did not involve a specific demographic group. Our research reveals that there is no tendency for users with similar characteristics to interact more. Overall, our study reveals how the war in Ukraine had a negative influence on daily conversations in the analyzed communities. This sheds light on how users responded to this significant event, providing insights into the dynamics of online discussions during events of global relevance.

Social media users drive the spread of misinformation online by sharing posts that include erroneous information or commenting on controversial topics with unsubstantiated arguments often in earnest. Work on echo chambers has suggested that users’ perspectives are reinforced through repeated interactions with like-minded peers, promoted by homophily and bias in information diffusion. Building on long-standing interest in the social bases of language and linguistic underpinnings of social behavior, this work explores how conversations around misinformation are mediated through language use. We compare a number of linguistic measures, e.g., in-/out-group cues, readability, and discourse connectives, within and across topics of conversation and user communities. Our findings reveal increased presence of group identity signals and processing fluency within echo chambers during discussions of misinformation. We discuss the specific character of these broader trends across topics and examine contextual influences.

Understanding defining features of conspiracy narratives presents a significant challenge. Existing research predominantly focuses on online platforms as alternative information sources, overlooking the self-reinforcing dynamics influencing content creators. This study addresses this gap by investigating the role of prior cognitive activation, specifically users’ engagement with conspiracy cues, in facilitating cognitive accessibility to such content. Central to our inquiry is the concept of activation burden, which refers to the cognitive effort required for individuals to engage with conspiracy narratives. We explore how features derived from evolutionary psychology, such as pattern recognition, detecting groups, or threat management, may contribute to the lowering of this activation burden, thereby fostering continued engagement with conspiratorial content.

To empirically examine these dynamics, we use data from Voat.co [26], a platform known for hosting de-platformed conspiracy-related discussions, sharing similarities in structure with Reddit. To characterize conspiracy content as multifaceted narratives, we investigate the utility of instruction-tuned Large Language Models (LLMs) for annotating text spans that represent various evolutionary facets of conspiracy features (N = 3, 384, between 2014-06-20 and 2020-12-23). Our findings highlight the self-reinforcing effects of cognitive activation, indicating that users responding to pattern, secrecy and threat show carry-over effects and persistence of these features. The results could contribute to the understanding of antecedents of conspiracy beliefs and platform moderation practices, enhanced by LLM-annotation.

Centralized social media platforms are currently experiencing a shift in user engagement, drawing attention to alternative paradigms like Decentralized Online Social Networks (DOSNs). The rising popularity of DOSNs finds its root in the accessibility of open-source software, enabling anyone to create a new instance (i.e., server) and participate in a decentralized network known as Fediverse. Despite this growing momentum, there has been a lack of studies addressing the effect of positive and negative interactions among instances within DOSNs. This work aims to fill this gap by presenting a preliminary examination of instances’ polarization in DOSNs, focusing on Mastodon — the most widely recognized decentralized social media platform, boasting over 10M users and nearly 20K instances to date. Our results suggest that polarization in the Fediverse emerges in unique ways, influenced by the desire to foster a federated environment between instances, also facilitating the isolation of instances that may pose potential risks to the Fediverse.

Decentralized Online Social Networks (DOSNs) are rising as a valid alternative to traditional centralized platforms like X (Twitter) and Facebook. Mastodon is to date the most widely recognized decentralized social media service. Thousands of servers have been deployed in the last few years due to the availability of open-source software which allows anyone to easily join the network of interconnected servers. Nonetheless, akin to other social media, Mastodon encompasses instances that host harmful or inappropriate content, which demands moderation. However, the decentralized nature of Mastodon servers poses novel challenges for content moderation. In this work, we explore the dynamics of decentralized moderation on Mastodon through the main tool offered to servers’ administrators, namely blocklisting servers to prevent users of an instance from interacting with the content of these servers. Our goal is to shed light on the main traits that characterize blocklisted instances on Mastodon and investigate the emergence of common blocklisting patterns toward specific groups of instances.

Common Crawl is a multi-petabyte longitudinal dataset containing over 100 billion web pages which is widely used as a source of language data for sequence model training and in web science research. Each of its constituent archives is on the order of 75TB in size. Using it for research, particularly longitudinal studies, which necessarily involve multiple archives, is therefore very expensive in terms of compute time and storage space and/or web bandwidth. Two new methods for mitigating this problem are presented here, based on exploiting and extending the much smaller (<200 gigabytes (GB) compressed) index which is available for each archive. By adding Last-Modified timestamps to the index we enable longitudinal exploration using only a single archive. By comparing the distribution of index features for each of the 100 segments into which archive is divided with their distribution over the whole archive, we have identified the least and most representative segments for a number of recent archives. Using this allows the segment(s) that are most representative of an archive to be used as proxies for the whole. We illustrate this approach in an analysis of changes in URI length over time, leading to an unanticipated insight into the how the creation of Web pages has changed over time.

YouTube is one of the most important platforms on the Internet. However, it is not just a singular destination: because YouTube videos may be embedded into any website, it is a systemically important platform for the entire web. Unfortunately, existing studies do not examine playable YouTube videos embedded around the web, instead focusing solely on on-platform engagement.

In this study we present the first comparison of on- and off-platform playable YouTube video exposure on the desktop web, based on data from a sample of n = 1133 U. S. residents from three cohorts: demographically representative users, heavy YouTube users, and users with high racial resentment. Our dataset includes the URLs of all playable YouTube videos encountered by participants for six months—on YouTube itself or embedded in another website—and the URLs where the videos were encountered. By analyzing this data, we find that less popular websites tend to embed YouTube videos in a disproportionately large percentage of their webpages and that individuals encounter more YouTube videos off-platform than on-platform. We also observe that most YouTube channels only receive exposure from either on- or off-platform sources, and that only-off-platform channels tend to get more exposure than only-on-platform ones. These results underscore the wide reach of YouTube as a service provider for the web and the limitations of studies focusing solely on on-platform activity.

Motivated by concerns about online misinformation and hate speech, we also examine how partisan websites embed playable YouTube videos and where videos from problematic channels appear around the web. We find that politically right-leaning websites tend to embed more videos from problematic YouTube channels than centrist or left-leaning websites, and that participants exposed to off-platform videos from problematic channels are significantly more inclined to browse towards on-platform videos from problematic channels.

We present a method for mapping Reddit communities that accounts for temporal shifts, using quantitative and qualitative analyses of clustering techniques to produce high-quality, stable, and meaningful maps for researchers, journalists and casual Reddit users. Building on previous work using community embeddings, we find that only a month of Reddit comments suffices to create snapshot embeddings that maintain quality while supporting insight into changes in Reddit communities over time. Comparing different clusterings of community embeddings with quantitative measures of quality and temporal stability, we describe properties of the models and what they tell us about the underlying Reddit data. Moreover, qualitative analysis of the resulting clusters illuminate which properties of clusterings are useful for analysis of Reddit communities. Although clusterings of subreddits have been used in many earlier works, we believe this is the first study to qualitatively analyze how these clusterings are perceived by social media researchers at a Reddit-wide scale.

Finally, we demonstrate how the temporal snapshots might be used in exploratory study. We are able to identify particularly stable communities during 2021-2022, such as the Reddit Public Access Network, as well as emerging communities, like one focused on NFT trading. This work informed the development of a webtool for exploring Reddit now available to the public at RedditMap.social.

In this paper we use for the first time a systematic approach in the study of harmonic centrality at a Web domain level, and gather a number of significant new findings about the Australian web. In particular, we explore the relationship between economic diversity at the firm level and the structure of the Web within the Australian domain space, using harmonic centrality as the main structural feature. The distribution of harmonic centrality values is analyzed over time, and we find that the distributions exhibit a consistent pattern across the different years. The observed distribution is well captured by a partition of the domain space into six clusters; the temporal movement of domain names across these six positions yields insights into the Australian Domain Space and exhibits correlations with other non-structural characteristics. From a more global perspective, we find a significant correlation between the median harmonic centrality of all domains in each OECD country and one measure of global trust, the WJP Rule of Law Index. Further investigation demonstrates that 35 countries in OECD share similar harmonic centrality distributions. The observed homogeneity in distribution presents a compelling avenue for exploration, potentially unveiling critical corporate, regional, or national insights.

Generative Artificial intelligence (GenAI) such as ChatGPT has elicited strong reactions from almost all stakeholders across the education system. Education-oriented and academic social media communities provide an important venue for these stakeholders to share experiences and exchange ideas about GenAI, which is constructive for developing human-centered policies. This study examines early user reactions to GenAI, consisting of 725 Reddit threads between 06/2022 and 05/2023. Through natural language processing (NLP) and content analysis, we observe an increasingly negative sentiment in the discussion and identify six main categories of student and faculty experiences of GenAI in education. These experiences reflect concerns about academic integrity and AI’s negative impact on the values of traditional education. Our analysis also highlights the tension and burden imposed by new technologies. Our findings suggest that dialogue between stakeholders in the education community is critical and can mitigate sources of tension between students and faculty.

This paper investigates how social media discussions evolved after the release and adoption of large language models by a wider public, with a focus on ChatGPT and the integration of a GPT model into Bing. The study aims to explore how social factors impact the way in which NLP technologies are perceived in social media posts. Using the official Twitter API, we collected a dataset of English and German tweets posted between November 30, 2022, and February 19, 2023. The study employs sentiment analysis and demographic prediction, the results reveal that tweets mentioning ‘Bing’ (related to the integration of a GPT model) were more likely to be negative compared to tweets about ‘ChatGPT’, with female users relatively more likely to express negative sentiment. The sentiment of tweets varied by language and whether an account belonged to an organization or not. This study provides insights into how social factors shape the discourse around NLP technologies on Twitter.

The kick-off of vaccination campaigns in Europe, starting in late December 2020, has been followed by the online spread of controversies and conspiracies surrounding vaccine validity and efficacy. We study Twitter discussions in three major European languages (Italian, German, and French) during the vaccination campaign. Moving beyond content analysis to explore the structural aspects of online discussions, our investigation includes an analysis of polarization and the potential formation of echo chambers, revealing nuanced behavioral and topical differences in user interactions across the analyzed countries. Notably, we identify strong anti- and pro-vaccine factions exhibiting heterogeneous temporal polarization patterns in different countries. Through a detailed examination of news-sharing sources, we uncover the widespread use of other media platforms like Telegram and YouTube for disseminating low-credibility information, indicating a concerning trend of diminishing news credibility over time. Our findings on Twitter discussions during the COVID-19 vaccination campaign in major European languages expose nuanced behavioral distinctions, revealing the profound impact of polarization and the emergence of distinct anti-vaccine and pro-vaccine advocates over time.

This paper focuses on the challenge of automatically detecting multimodal fake news on social media. Although multimodal fake news classifiers exist, we show that prior works fail to reflect certain real-world practicalities. For example, news captions often contain highly irrelevant information that introduces noise to the overall message contained within the post. Existing classifiers do not properly address this, resulting in misclassifications. To address this limitation and suppress noise, we propose GatedVAE (Gated Variational AutoEncoder), which enables VAE with the gating mechanism. Experimental results demonstrate the efficacy of our approach: GatedVAE is able to suppress noise and learn an effective multimodal representation. It outperforms state-of-the-art models by 3.7% and 2.4% (F1) on Twitter and Weibo datasets, respectively. Our ablation study highlights the importance of the gating mechanism and the methods we adopt to alleviate overfitting. We further show that, in addition to dynamically controlling the pass of noisy input, the gate also helps to uncover modality importance in multimodal fake news detection.

With the primary goal of raising readers’ awareness of misinformation phenomena, extensive efforts have been made by both academic institutions and independent organizations to develop methodologies for assessing the trustworthiness of online news publishers. Unfortunately, existing approaches are costly and face critical scalability challenges. This study presents a novel framework for assessing the trustworthiness of online news publishers using user interactions on social media platforms. The proposed methodology provides a versatile solution that serves the dual purpose of i) identifying verifiable online publishers and ii) automatically performing an initial estimation of the trustworthiness of previously unclassified online news outlets.

Online misinformation poses a global risk with harmful implications for society. Ordinary social media users are known to actively reply to misinformation posts with counter-misinformation messages, which is shown to be effective in containing the spread of misinformation. Such a practice is defined as “social correction”. Nevertheless, it remains unknown how users respond to social correction in real-world scenarios, especially, will it have a corrective or backfire effect on users. Investigating this research question is pivotal for developing and refining strategies that maximize the efficacy of social correction initiatives.

To fill this gap, we conduct an in-depth study to characterize and predict the user response to social correction in a data-driven manner through the lens of X (Formerly Twitter), where the user response is instantiated as the reply that is written toward a counter-misinformation message. Particularly, we first create a novel dataset with 55,549 triples of misinformation tweets, counter-misinformation replies, and responses to counter-misinformation replies, and then curate a taxonomy to illustrate different kinds of user responses. Next, fine-grained statistical analysis of reply linguistic and engagement features as well as repliers’ user attributes is conducted to illustrate the characteristics that are significant in determining whether a reply will have a corrective or backfire effect. Finally, we build a user response prediction model to identify whether a social correction will be corrective, neutral, or have a backfire effect, which achieves a promising F1 score of 0.816. Our work enables stakeholders to monitor and predict user responses effectively, thus guiding the use of social correction to maximize their corrective impact and minimize backfire effects. The code and data is accessible on https://github.com/claws-lab/response-to-social-correction.

Recommendation algorithms (RAs) have been pointed out as one of the major culprits of misinformation spreading in the digital sphere.1 However, it is still unclear how these algorithms propagate misinformation, e.g., which particular recommendation approaches are more prone to suggest misinforming items, or which internal parameters of the algorithms could be influencing more on their misinformation propagation capacity. Motivated by this fact, in this work, we present an analysis of the effect of some of the most popular recommendation algorithms on the spread of misinformation on Twitter (X). A set of guidelines on how to adapt these algorithms is provided based on such analysis and a comprehensive review of the research literature.

This study investigates the ranking of problematic content and fact-checks of that content in Google Web Search results, examining their competition. The analysis is based on over 825 URLs extracted from Google Search Engine results pages (SERP) using 50 thematic keywords derived from fact-checked content. It reports findings on both the rankings as well as the levels of SEO optimisation of problematic content and the fact-checks. Our investigation found that fact-checks enjoy greater visibility in Google Web Search compared to the articles they seek to correct, both in terms of frequency of appearance and their placement within the SERP rankings. Specifically, our study shows fact-checks rank higher than problematic content across five topical keywords groups, Covid-19, climate change, the war in Ukraine, U.S. liberals and U.S. elections, except in contested stories related to the war in Ukraine, where articles about U.S. bio-labs share equal prominence with their corresponding fact-checks. The findings imply Google moderation effects, as fact-checking content is more prominent given (nearly) equal levels of optimisation. It also implies that fact-checks are generally more prominent for audiences searching for problematic content, though both often appear in the same SERP. Navigational queries (e.g., searching for the name of a source and that content) reduce moderation effects.

The viral phenomenon is present in several contexts, combining the advantages of streaming platforms and other social networks. Music is no exception. Viral songs are widely shared in a short amount of time, and they may become successful by reaching the top of the charts with millions of streams and digital sales. However, not all songs that go viral become hits, as the sharing process is not enough to be converted into streams. In this work, we analyze viral and hit songs as two different yet interconnected aspects of music popularity. Specifically, we aim to uncover factors that are relevant to distinguishing viral from hit songs. We evaluate three research hypotheses related to musical features, and the results reveal that considering only acoustic and other intrinsic features is not enough; and extrinsic features (e.g., artists’ genre and the time from the release to the first chart entry) are essential in achieving such a goal. Moreover, artist-related and temporal features are among the most relevant indicators to differentiate hit and viral songs. Overall, our findings offer relevant insights for understanding the dynamics of music consumption and its sharing on online platforms, as we reveal important factors that describe what makes a viral song and what differentiates it from a hit.

The rise of on-demand music streaming platforms and novel recommendation algorithms have brought a transformative shift in music listening, where users have an effectively endless supply of new music to discover. This study aims to understand novel music streaming patterns, operationalizing novel music as new music releases that are new to everyone, from the perspective of genres. Leveraging tracks, users, and streaming data from Spotify, we empirically analyze streaming patterns of 282K new music releases. We find that new releases in genres that often serve functional purposes, such as classical music for relaxation, are consumed less. Surprisingly, new releases in Pop generally do not exhibit high consumption rates despite being characterized as popular music. From examining 1 million users’ historical genre preferences, we observe that users’ new release preferences are distinct from their overall music preferences, although past genre affinities and proclivity for newer content can predict new music consumption. Our findings present important implications for new music recommendation strategies on music streaming platforms and broader contributions to understanding the dynamics of music discovery.

Music is intertwined with geography as part of the cultural fabric that transforms a physical space into what we consider a “place." But in an increasingly online world, where music’s distribution has shifted massively towards digital streaming on global platforms, what role does geography now play in shaping peoples’ music consumption? Here, we employ a multi-part, mixed-methods study of “local" music, exploring its current definition as well as exploring its potential role in online music recommender systems. We present, first, findings from a qualitative study designed to identify themes in how listeners and artists defined local music across three international locations. Second, we present results of a quantitative analysis that operationalizes this definition and investigates the impact of surfacing local context in a real-world recommendation setting, conducted in one location. Together, our results illustrate that “local" continues to play a crucial role in shaping music’s enjoyment and represents an important mechanism for facilitating the discovery of lesser-known artists in online algorithmic recommendations.

YouTube introduced the Shorts video format in 2021, allowing users to upload short videos that are prominently displayed on its website and app. Despite having such a large visual footprint, there are no studies to date that have looked at the impact Shorts introduction had on the production and consumption of content on YouTube. This paper presents the first comparative analysis of YouTube Shorts versus regular videos with respect to user engagement (i.e., views, likes, and comments), content creation frequency and video categories. We collected a dataset containing information about 70k channels that posted at least one Short, and we analyzed the metadata of all the videos (9.9M Shorts and 6.9M regular videos) they uploaded between January 2021 and December 2022, spanning a two-year period including the introduction of Shorts. Our longitudinal analysis shows that content creators consistently increased the frequency of Shorts production over this period, especially for newly-created channels, which surpassed that of regular videos. We also observe that Shorts target mostly entertainment categories, while regular videos cover a wide variety of categories. In general, Shorts attract more views and likes per view than regular videos, but attract less comments per view. However, Shorts do not outperform regular videos in the education and political categories as much as they do in other categories. Our study contributes to understanding social media dynamics, to quantifying the spread of short-form content, and to motivating future research on its impact on society.

Despite Instagram being an integral part of many people’s lives, it is relatively less studied than many other platforms (e.g., Twitter and Facebook). Furthermore, despite offering diverse content formats for user expression and interaction, prior works have not studied the temporal dynamics of user engagement across albums, photos, and videos. To address this gap, we present a pioneering temporal comparative analysis that unveils nuanced patterns in user interactions across content types. Our analysis sheds light on interaction longevity and disparities among album, photo, and video engagement. Additionally, it offers empirical comparisons through statistical tests, examines contributing factors such as post and uploader characteristics, and analyzes content composition’s impact on user engagement. The findings reveal distinct temporal engagement patterns. Despite initial spikes in interactions post-upload, albums exhibit somewhat more sustained interest, while photos and videos have shorter engagement lifespans. Moreover, a consistent trend between shallow (likes) and deep (comments) interactions persists across content types. Notably, concise content, characterized by shorter descriptions and minimal hashtags/mentions, consistently drives higher engagement, emphasizing its relevance across all content formats. These insights deepen comprehension of temporal nuances in user engagement on Instagram, offering valuable guidance for content creators and marketers to tailor strategies that evoke immediate and sustained user interest.

Generative AI for the creation of images is becoming a staple in the toolkit of digital artists and visual designers. The interaction with these systems is mediated by prompting, a process in which users write a short text to describe the desired image’s content and style. The study of prompts offers an unprecedented opportunity to gain insight into the process of human creativity. Yet, our understanding of how people use them remains limited. We analyze more than 145,000 prompts from the logs of two Generative AI platforms (Stable Diffusion and Pick-a-Pic) to shed light on how people explore new concepts over time, and how their exploration might be influenced by different design choices in human-computer interfaces to Generative AI. We find that users exhibit a tendency towards exploration of new topics over exploitation of concepts visited previously. However, a comparative analysis of the two platforms, which differ both in scope and functionalities, reveals some stark differences. Features diverting user focus from prompting and providing instead shortcuts for quickly generating image variants are associated with a considerable reduction in both exploration of novel concepts and detail in the submitted prompts. These results carry direct implications for the design of human interfaces to Generative AI and raise new questions regarding how the process of prompting should be aided in ways that best support creativity.

While minimising false negatives in hate speech classification remains an important goal in order to reduce discrimination and increase fairness for online communities, there is a growing need to produce models that are sensitive to nuanced language use. This is particularly true for terms that may be considered hateful in certain contexts, but not others. The LGBTQ+ community has long faced stigmatisation and hate, which continues to be the case online. There has been a rise in appreciation and understanding of this community’s use of “mock impoliteness" and the reclaiming of language that has traditionally been used derogatorily against them. Reclaimed language in particular presents a challenge in the field of hate-speech detection. As a first-of-its-kind study looking into the impact of reclaimed language on hate speech detection models, we create a novel dataset, Reclaimed Hate Speech Dataset (RHSD), which enables investigation into the phenomenon. Through the use of a state-of-the-art hate speech detection model, we demonstrate that models may inadvertently discriminate against the LGBTQ+ community’s reclaimed language use through misclassifying such content as hateful. As a result, there is a risk of compounding discrimination against this population through restricting their language use and self-expression. In response to this issue, we produce a fine-tuned hate-speech detection model which aims to minimise false positive classifications of reclaimed language. By creating and publishing the first dataset that focuses on reclaimed language and investigating its impact on hate speech detection models, our research highlights the importance of semantically aware approaches to hate-speech detection that are not overly reliant on individual words or phrases associated with hate. We thus establish a benchmark methodology for further investigation into reclaimed language, that promises to support marginalised groups, taking into account the intersectional nature of their discourse.

NB: Readers should be advised that this paper contains use of and reference to racial, homophobic and transphobic slurs which they may find triggering. References to sentences containing such slurs are purely for illustration purposes and in no way reflect the author’s attitudes or opinions.

Online communities are groups of people who interact primarily via the Internet, often sharing common interests. Some of these groups, particularly supporters of Q who created the far-right conspiracy theory known as QAnon, are highly toxic and controversial. These communities are often banned from various mainstream online social networks due to their controversy. This study examines the deplatforming and subsequent migrations of QAnon adherents, following a two-step process. We analyze Reddit data, finding that users opt for Voat as an alternative following the Reddit bans, particularly influenced by Q’s postings on 4chan. Subsequently, upon Voat’s shutdown announcement, we observe users recommending Poal. Among several insights, we compare the effects of abrupt permanent bans and announced shutdowns on the migration patterns of these conspiracists. Specifically, we find that almost half of Poal’s active users are Voat migrants who registered after the shutdown was announced. This contradicts the patterns observed after Reddit bans, suggesting that advance warning can facilitate more coordinated migrations. Lastly, our research uncovers evidence of discussions and planning related to the January 6th, 2021, attack on the US Capitol, which emerged shortly after Voat’s shutdown, predominantly on Poal. This underscores the continued activity of the conspiracy, albeit at a diminished scale due to various bans and a shutdown, while also exposing Poal as a platform that hosts dangerous individuals.

Incels are an extremist online community of men who believe in an ideology rooted in misogyny, racism, the glorification of violence, and dehumanization. In their online forums, they use an extensive, evolving cryptolect – a set of ingroup terms that have meaning within the group, reflect the ideology, demonstrate membership in the community, and are difficult for outsiders to understand. This paper presents a lexicon with terms and definitions for common incel root words, prefixes, and affixes. The lexicon is text-based for use in automated analysis and is derived via a Qualitative Content Analysis of the most frequent incel words, their structure, and their meaning on five of the most active incel communities from 2016 to 2023. This lexicon will support future work examining radicalization and deradicalization/disengagement within the community.

Web-based content analysis tasks, such as labeling toxicity, misinformation, or spam often rely on machine learning models to achieve cost and scale efficiencies. As these models impact real human lives, ensuring accuracy and fairness of such models is critical. However, maintaining the performance of these models over time can be challenging due to the temporal shifts in the application context and the sub-populations represented. Furthermore, there is often a delay in obtaining human expert labels for the raw data, which hinders the timely adaptation and safe deployment of the models. To overcome these challenges, we propose a novel approach that anticipates future distributions of data, especially in settings where unlabeled data becomes available earlier than the labels to estimate the future distribution of labels per sub-population and adapt the model preemptively. We evaluate our approach using multiple temporally-shifting datasets and consider bias based on racial, political, and demographic identities. We find that the proposed approach yields promising performance with respect to both accuracy and fairness. Our paper contributes to the web science literature by proposing a novel method for enhancing the quality and equity of web-based content analysis using machine learning. Experimental code and datasets are publicly available at https://github.com/Behavioral-Informatics-Lab/FAIRCAST.

Search engines are commonly used for online political information seeking. Yet, it remains unclear how search query suggestions for political searches that reflect the latent interest of internet users vary across countries and over time. We provide a systematic analysis of Google search engine query suggestions for European and national politicians. Using an original dataset of search query suggestions for European politicians collected in ten countries, we find that query suggestions are less stable over time in politicians’ countries of origin, when the politicians hold a supranational role, and for female politicians. Moreover, query suggestions for political leaders and male politicians are more similar across countries. We conclude by discussing possible future directions for studying information search about European politicians in online search.

The web is a major source for news and information. Yet, news can perpetuate and amplify biases and stereotypes. Prior work has shown that training static word embeddings can expose such biases. In this short paper, we apply both a conventional Word2Vec approach as well as a more modern BERT-based approach to a large corpus of Dutch news. We demonstrate that both methods expose ethnic biases in the news corpus. We also show that the biases in the news corpus are considerably stronger than the biases in the transformer model itself.

In this study we investigate gender bias within journalist politician interactions in India—a country in global south. Previous work identified gender bias in online political discourse, but they focused on global north and politician-general public interaction. However, under-studied journalist-politician interactions can significantly affect public sentiment and help set gender-biased social norms in collectivistic cultures like India, motivating this work.

Specifically, we curated a gender-balanced set of the 100 most-followed Indian journalists on Twitter and the 100 most-followed politicians. Then we collected 21,188 unique tweets posted by these journalists that mentioned these politicians. Our analysis revealed that there is a significant gender bias—the frequency with which journalists mention male politicians vs. how frequently they mention female politicians is statistically significantly different (p < < 0.05). In fact, median tweets from female journalists mentioning female politicians received ten times fewer likes than median tweets from female journalists mentioning male politicians. However our emotion score analysis, topic modeling analysis and semantic similarity analysis did not reveal any significant difference between the journalists’ tweet content mentioning male politicians and those mentioning female politicians. We identify a potential reason—the high discrepancy within the number of popular male vs. female Indian politicians. Our work hint at need for platform-mediated gender diversity in the online Indian political discourse.

Social media usage has been shown to have both positive and negative consequences for users’ mental health. Several studies indicated that peer feedback plays an important role in the relationship between social media use and mental health. In this research, we analyse the impact of receiving online feedback on users’ emotional experience, social connectedness and self-esteem. In an experimental study, we let users interact with others on a Facebook-like system over the course of a week while controlling for the amount of positive reactions they receive from their peers. We find that experiencing little to no reaction from others does not only elicit negative emotions and stress amongst users, but also induces low levels of self-esteem. In contrast, receiving much positive online feedback, evokes feelings of social connectedness and reduces overall loneliness. On a societal level, our study can help to better understand the mechanisms through which social media use impacts mental health in a positive or negative way. On a methodological level, we provide a new open-source tool for designing and conducting social media experiments.

In the global discourse on women’s reproductive rights, both miscarriage and abortion can be highly charged topics of discussion and are both at risk of being affected by stigmatisation and legal measures. Analysing the discussions on social media around these topics can help us gauge the thoughts and feelings of the general public, providing valuable insights into prevailing sentiments and perspectives on matters related to reproductive rights. The main objective of this research is to gain a deeper understanding of opinions and sentiments on Twitter (now "X") surrounding miscarriage and abortion, looking at who is posting, the sentiment of the posts, and how this impacts the post’s popularity. To do this, we gather English tweets from Twitter using both keywords.

In 2021, the Reddit community of WallStreetBets (WSB) started making headlines in the mainstream media as a source of risky investment ideas called YOLOs, and meme stocks such as GameStop. The community has become infamous for its use of vulgar language, memes, and unorthodox investment strategies. While previous research has explored the community in terms of language and social dynamics, this work sheds light on the extraction of predictive value from investment advice shared on WSB to investigate Reddit’s collective intelligence. We evaluate more than 1.6 million posts contributed over 4.5 years, extracting thousands of investment recommendations, which are matched and benchmarked with corresponding stock market data. To analyse and assess their value, we investigate multiple machine learning models, evaluating their effectiveness on multiple time windows under different market conditions, and exploring their latent spaces. When predicting the success of a new post, we view it within the context of the WSB community’s discussions and the stock market data of the same day. Our best model’s predictive performance on stock market data outperforms the S&P 500 index and other traditional investment strategies by a significant margin. We conclude that amateur retail traders posting on WSB can act as an intelligent crowd and constitute a valuable source of investment advice. Our findings yield generalizable insights on how the collective intelligence of online communities can be extracted and utilized.

Cryptocurrency is a fast-moving space, with a continuous influx of new projects every year. However, an increasing number of incidents in the space, such as hacks and security breaches, threaten the growth of the community and the development of technology. This dynamic and often tumultuous landscape is vividly mirrored and shaped by discussions within “Crypto Twitter,” a key digital arena where investors, enthusiasts, and skeptics converge, revealing real-time sentiments and trends through social media interactions. We present our analysis on a Twitter dataset collected during a formative period of the cryptocurrency landscape. We collected 40 million tweets using keywords related to cryptocurrency and performed a nuanced analysis that involved grouping the tweets by semantic similarity and constructing a tweet and user network. We used sentence-level embeddings and autoencoders to create K-means clusters of tweets. We identified six groups of tweets and their topics to examine different cryptocurrency-related interests and the change in sentiment over time. For example, we identified different groups of tweets demonstrating coordinated behavior in the market or expressing distrust in centralized cryptocurrency exchanges. Moreover, we discovered sentiment indicators that point to real-life incidents in the crypto world, such as the FTX incident of November 2022. We also constructed and analyzed different networks of tweets and users in our dataset by considering the reply and quote relationships and analyzed the largest components of each network. Our networks reveal a structure of bot activity in Crypto Twitter and suggest that they can be detected and handled using a network-based approach. Our work sheds light on the potential of social media signals to detect and understand crypto events, benefiting investors, regulators, and curious observers alike, as well as the potential for bot detection in Crypto Twitter using a network-based approach.

The landscape of human interaction has undergone a profound transformation since the advent of Online Social Networks (OSNs). Not only are they changing interpersonal dynamics, but they are also redefining the way businesses, political figures, and media organizations engage with the broader population. In today’s digital landscape, OSNs have spawned a new class of social media influencers who play a crucial role in shaping opinion. These influencers actively compete within social media to seize users’ attention. Through these targeted efforts, influencers seek to captivate users and build a loyal and engaged fan base, solidifying their position as an authoritative voice in the digital world. In this work, we develop a game-theoretic model for the interactions between users and influencers, where the latter compete to maximize their impact on the population’s opinions. The goal of this work is twofold: first, we formalize the problem of maximizing social media impact and study the structure of the optimal solution. Then, taking inspiration from the optimal strategy, we develop a game with two opposing players trying to maximize their influence on users’ opinions, for which we characterize the Nash equilibria in pure strategy. The model allows us to evaluate the impact of influencer differences and user population characteristics. In addition, we study the effect of the speed at which user popularity evolves in such a competitive environment. The proposed model proves valuable for brand competition, marketing campaigns, and the ever-evolving political arena.

Our world is shaped by events of various complexity. This includes both small-scale local events like local farmer markets and large complex events like political and military conflicts. The latter are typically not observed directly but through the lenses of intermediaries like newspapers or social media. In other words, we do not witness the unfolding of such events directly but are confronted with narratives surrounding them. Such narratives capture different aspects of a complex event and may also differ with respect to the narrator. Thus, they provide a rich semantics concerning real-world events. In this paper, we show how narratives concerning complex events can be constructed and utilized. We provide a formal representation of narratives based on recursive nodes to represent multiple levels of detail and discuss how narratives can be bound to event-centric knowledge graphs. Additionally, we provide an algorithm based on incremental prompting techniques that mines such narratives from texts to account for different perspectives on complex events. Finally, we show the effectiveness and future research directions in a proof of concept.

Addressing the climate crisis calls for global efforts to mitigate greenhouse gas emissions, with effective mitigation efforts depending on an informed public with a high degree of climate change awareness. This study examines global engagement with climate change and related concepts through an analysis of around 517 Million Wikipedia pageviews of 3965 items from WikiProject Climate Change across 213 countries in the years 2017 to 2022. We take advantage of Wikimedia Foundation’s differentially-private daily pageview dataset, which makes it possible to study Wikipedia viewing behavior in a language edition agnostic way and on a per-country basis. Temporal analysis reveals a stagnant engagement with climate change articles, contrary to societal trends, possibly due to the attitude-behavior gap. We also found substantial regional differences, with countries from the global north displaying greater traffic compared to the global south. Specific events, notably Greta Thunberg’s speech at the UN climate summit in 2019, drive peaks in climate change engagement, highlighting the social dimension and influence of prominent figures in climate change information seeking. However, causal time series analyses show that events like these do not lead to long-lasting increased traffic.

Public discourse on critical issues such as climate change is progressively shifting to social media platforms that prioritize short-form video content. Content creators acting on those platforms play a pivotal role in shaping the discourse, yet the dynamics of communication and audience reactions across platforms remain underexplored. To improve our understanding of this transition, we studied the video content produced by 21 prominent YouTube creators who have expanded their influence to TikTok as information disseminators. Using dictionary-based tools and BERT-based embeddings, we analyzed the transcripts of nearly 7k climate-related videos across both platforms and the 574k comments they received. We found that, when publishing on TikTok, creators use a more emotionally resonant, self-referential, and action-oriented language compared to YouTube. We also observed a strong semantic alignment between videos and comments, with creators who excel at diversifying their TikTok content from YouTube typically receiving responses that more closely align with their produced content. This suggests that tailored communication strategies hold greater promise in directing public discussion towards desired topics, which bears implications for the design of effective climate communication campaigns.