ABSTRACT. We study the migration of researchers in Germany in 1996-2020 using an exhaustive set of 8 million Scopus publications. Not only emigrant researchers outnumber immigrant researchers, but also, they outperform immigrants in average annual citations. Substantial gender disparities exist in almost all disciplines to which migration has provided some counterbalance.

ABSTRACT. We investigated the unfolding emotional consequences of social sharing during a crisis. By combining a large scale twitter study with a event diary survey we were able to find consistent evidence for the adaptive process of social sharing. The incremental value of the multimethod approach will also be discussed.

ABSTRACT. Recently, WhatsApp is turning into a breeding ground for hatred in India. Here, we operationalise "fear speech", which incites fear about a community. We annotate 5000 posts containing fear/non-fear speech from Whatsapp groups. We found topics and emojis used to demonise a community. Fear speech detection remains an open challenge.

ABSTRACT. Through a multi-modal framework, this paper explores the presence and characteristics of Chinese government content on video-sharing social media by 50,813 videos collected from the Trending page of Douyin. The methods and substantive findings contribute to an emerging literature in communication on the computational analysis of video as data.

ABSTRACT. This paper introduces a novel dataset about popular music and employs it to study cultural influence. Using computational methods, we classify the lyrics and find frames in line with previous literature about inequalities. Merging this dataset with existing survey data, we find an association between public frames and beliefs.

ABSTRACT. Our work highlights the challenges encountered in research with digital trace data, inspired from survey methodology. Since many new challenges arise due to idiosyncrasies of the big and heterogenous data encountered in digital spaces, we develop the Total Error Framework for Digital Traces of Online Human Behavior (TED-On).

ABSTRACT. We introduce a new computational framework for studying intimacy in language with accompanying datasets and NLP models for predicting the intimacy level of questions. We demonstrate how individuals modulate their intimacy to match social norms around gender, social distance, and anonymity, each validating key findings from social psychology. Website: https://blablablab.si.umich.edu/projects/intimacy/

ABSTRACT. This work offers a graph-based view on the discussions between SM contributors and to retrieve popular patterns on online conversations using network-based analysis. The proposed solution consists of three main stages: intent analysis, network generation, and pattern identification. We tested the proposed methodology on a real SM game challenge event.

ABSTRACT. We have designed and implemented a web data extraction framework for live monitoring and archiving of discussions from the 4chan online community. Thanks to the deployment of an advanced version of the OxPath crawler, we were able to track the evolution of the /pol/ bilboard with higher level of precision.

ABSTRACT. Recently, YouTube's outsize influence has sparked concerns that its recommendation algorithm systematically directs users to radical content. We investigate these concerns with large scale longitudinal data of individuals’ browsing behavior. We find evidence for echo chambers of radical content, but no evidence that they are caused by YouTube recommendations.

ABSTRACT. We present a method to infer ideological spaces latent to online social networks. In these spaces, dimensions are indicators for attitudes towards issues of the public debate. With this method, we are able to embed entities from different social platforms and follow the ideological trajectory of partisan public Facebook groups.

ABSTRACT. We used the deep learning method, the bidirectional encoder representations from transformers (BERT) language model, to analyze news bulletin board comments under the COVID-19 epidemic, and found that the valence frame and topic frame of news articles had a significant impact on the expression of anti-Chinese sentiment.

ABSTRACT. By analyzing over 670,000 teams in 45 years, we show that teams with unexpected collaboration between institutions are associated with higher impact. We offer a new perspective (unexpectedness) on the salience of team assembly for scientific success beyond the well-known factors traditionally investigated (e.g., team size and number of institutions).

ABSTRACT. Understanding the dynamics of interdisciplinarity has played a central role in current debates of the Science of Science. In this work, we leverage large-scale bibliometric data (Microsoft Academic Graph) and quantify the temporal patterns of interdisciplinarity in scientists' careers.

ABSTRACT. In this paper, we conduct a systematic literature review on computational mental health research in the context of computer-mediated communication, focusing both on theoretical and methodological challenges. We aim to describe the field, reveal common data sources and to explore the scope of cutting-edge computational methods in this research area.

ABSTRACT. The scholars' trajectories of foraging for research in knowledge space are characterized by subfield constraint, frequency bias, proximity bias, and the relationship between foraging scope and academic performance shows the knowledge boundary effect.

ABSTRACT. The collaborative nature of code development if found to be linked to the trading activity of cryptocurrencies, showing that, in this ecosystem, transparency is a network property.

ABSTRACT. We use large-scale behavioral logs from the KidZania theme park (a child-sized replica of a real city) to analyze girls’ and boys’ preferences for occupations and find that there are gender preferences in occupations that increase with age.

ABSTRACT. We employ a socio-metric survey in a small portion of a large consumer goods company to identify positive and negative ties, then use these data in addition to anonymized email conversation data and HR-supplied attribute data to build a multi-stage deep-learning model to predict tie valence throughout the entire organization.

ABSTRACT. In this work, we investigate home-work locations and mutual followership ties of Twitter users from the top 50 metropolitan areas of the United States. We find that despite the heterogeneity of spatial structures in cities, above median commuting reduces the income assortativity of social networks by 30% on average.

ABSTRACT. In a digital society, machines can influence cultural evolution not only by spreading human behaviour but also by providing novel behaviours themselves. However, do these behaviours stick within human culture? We ask whether machines with complementary biases could boost cultural evolution in a lab experiment when human biases are suboptimal.

ABSTRACT. In a large-scale, pre-registered experiment on informal political communication on the topic of government redistribution of wealth, we bridge sources and recipients of political influence through their non-political commonalities. We show that consensus increases, polarization is reduced, and feeling close to the source of the message greatly enhance this result.

ABSTRACT. In this study we use a large-scale, international experiment among Airbnb users to answer the question how participating in interactions on Airbnb affects cross-national trust, and to examine whether the effect of prior experience depends on several contextual factors.

ABSTRACT. Our study examines the effectiveness of specific crowd signals (i.e., the momentum and variation of contributions) in a mock crowdfunding setting. We investigate whether high crowd signals influence people’s decisions to contribute to a crowdfunding project and whether contributions are higher when people are more susceptible to social influence.

ABSTRACT. This work aims at characterizing patterns of the nonmedical consumption of opioids in the US, as discussed by users on Reddit, with a focus on routes of administration and drug tampering. We learned the platform-specific terminology using word embedding, finding quantitative evidence of unconventional nonmedical consumption behaviors.

ABSTRACT. We calibrate an agent based model using empirical data of SARS-CoV-2 outbreaks in Austrian schools. We use the model to assess the effectiveness of non-pharmaceutical intervention measures in different school types. We find that different measure combinations are necessary for different school types and make a controlled re-opening possible.

ABSTRACT. In this work, we explore the connection between educational attainment, as a proxy for cultural capital, and food purchases, as a proxy for food consumption. Our results consistently confirm the association between a higher educational attainment and a healthier diet, even when controlling for spatial correlation.

ABSTRACT. Cities are the innovation centers of the US economy, but disruptions from technology may exclude workers and inhibit a middle class. Urban policy must promote the jobs and skills that increase pay and employment, and make their economy resilient. This study describes a framework for achieving such predictive capability.

ABSTRACT. Residential displacement occurs when individuals are forced out of an area as costs rise. If identified early, policy can mitigate the process and its negative impacts. Motivated by this, and using a new dataset of home transactions in a U.S. city, we study the predictability of displacement.

ABSTRACT. Semantic social networks are a way to encode large ethnographic corpora as structured data, and express them as networks. We draw on methods from network mathematics and qualitative research to compare different techniques to reduce the network and make them amenable to visual analysis.

ABSTRACT. Virtual meetings are critical for remote work. In-meeting multitasking is closely linked to people's productivity and wellbeing. Here we present what we believe is the most comprehensive study of remote meeting multitasking behavior through an analysis of a large-scale telemetry dataset and a 715-person diary study.

ABSTRACT. Focusing on Reddit, we estimate that nearly half of all political talk takes place in subreddits which host political content less than 25% of the time. Thus, cumulative political talk in nonpolitical spaces is abundant. These conversations in nonpolitical spaces are less toxic than such interactions in political spaces.

ABSTRACT. When analyzing relationship between two variables, the use of imperfect measurements could lead to substantial distortion of results. We demonstrate that this problem is not purely theoretical and arise naturally in a typical CSS setup. We hope that this cautionary tale would raise awareness of the problem in our community.

ABSTRACT. Using large-scale textual conversation data collected from multiple domains, we show that ten different types of social dimensions can be inferred from conversations using machine learning approaches, and further show that the inferred dimensions reflect relationships between people, changes that organizations experience, and societal outcomes of real-world communities.

ABSTRACT. This project studies China's social media campaign to examine how online propaganda could shape public discourse. Findings reveal that, during the campaign, the public was influenced by the official propagandist and started to adopt official discourses. The result expands our conceptualization of information control beyond censorship, trolling, and disinformation.

ABSTRACT. Retracted papers are harmful to the public if they are exposed to the flawed findings. Here we study collective attention on various online platforms ranging from social media to knowledge repositories for about 4,000 papers retracted in the last four decades to measure how much retracted papers diffuse online.

ABSTRACT. We explore the existence of political acrophily, which is the tendency to associate with others who hold more politically extreme views than one’s own. Analyzing a large retweet network of users, we show that both conservatives and liberals prefer to retweet content produced by more politically extreme users than them.

ABSTRACT. We analysed more than 100 million Twitter messages posted worldwide during the early stages of epidemic spread across countries and classified the reliability of the news being circulated. We developed an Infodemic Risk Index to capture the magnitude of exposure to unreliable news across countries.

ABSTRACT. This study applies dynamic word embedding techniques on millions of newspaper and patent documents to construct a dynamic landscape of business discourse across 45 years. This allows us to observe how business elements are recombined within a given new venture, and how this creative recombination influences a start-up’s future success.

ABSTRACT. We illustrate a mathematical connection between the neural embedding and the gravity model of mobility and, using three human trajectory datasets, demonstrate that a neural embedding encodes a multi-faceted structure of human mobility. Also, we show that embeddings of scientific organizations uncover cultural and linguistic relations, and even academic prestige.

ABSTRACT. We show that inherent biases in random walks have a profound impact on graph embeddings based on random walks such as DeepWalk and node2vec, and propose residual2vec to correct them, which produces embeddings that are more useful for link prediction and community detection and reveal a salient relationship of nodes.

ABSTRACT. Our research quantifies the differences between North American and European researcher’s collaboration networks , and further examines how such differences impact scientific productivity at both sides of the Atlantic. We contain that understanding differences in collaboration networks between regions is crucial for developing policies to overcome barriers to scientific progress

ABSTRACT. We examine the structure of scientific collaborations and strategic coalitions in two regions in Germany with specific historical, geographic and institutional backgrounds. A densely populated metropolitan region with high political influence, i.e., Berlin and a geographically dispersed region consisting of three cities in central Germany, i.e., Halle, Jena and Leipzig.

ABSTRACT. To understand the declining local news industry's impact on the diversity of news stories published in the United States, we measure the proportion of articles that are duplicated between different newspapers and present a novel technique that exploits shared structure in article URLs to automatically extract networks of media consolidation.

ABSTRACT. This study shows that online social connections of people in poor neighborhoods are more spatially concentrated and structurally cohesive than the network of people living in better-off areas of the top 50 metropolitan areas of the US.

ABSTRACT. We explore the structure of meanings in human language through the analysis of colexification networks, developing methods to automatically improve affective science resources. We prove that affective meaning is encoded in colexification networks and deploy them to improve state of the art methods to infer the affective ratings of words.

ABSTRACT. Throughout history, a relatively small number of researchers have published both journal papers and patents. These authors are typically referred to as author-inventors. In this study, we focused on author-inventors to investigate the effect that publishing their first patent has on scientists’ careers.

ABSTRACT. We test the assumption that juries make consistent decisions using an online experiment: counter to expectation, we find that deliberative juries are equally consistent as individuals, but aggregating groups' perspectives without deliberation erodes consistency.

ABSTRACT. Focusing on sequential decision-making, we design a novel machine-learning algorithm that can convey its insights to humans in the form of interpretable "tips". We evaluate our approach through a series of randomized controlled experiments where participants manage a virtual kitchen and show that our tips can significantly improve human performance.

ABSTRACT. I run a field experiment that collects comprehensive data on the time usage of participants. I randomize restrictions to several prominent social media applications on participant's phones and measure how they reallocate their time usage. I use the measured substitution patterns to guide market definitions in zero-price attention markets.

ABSTRACT. A pre-registered web experiment examined how reply and reaction (upvote and downvote) -- two of social media's most frequently employed interactive features -- affect the communication dynamics of political discussions on a messaging platform custom-designed by the researchers.

ABSTRACT. Non-influenza respiratory viruses, including common coronaviruses, contribute to the burden of the flu-season. Using epidemiological surveillance data, climatic data, and Google search trends, we are gaining insights in the seasonal dynamics of NIRVs, in the possible impacts of climate, as well as in nowcasting models for the flu-NIRVs seasonal epidemic.

ABSTRACT. It has been shown that being part of more online groups and actively participating in them can help addicts to maintain their recovery. In this study, we further disentangle the impact of different types of online groups, those that are beneficial to maintaining recovery and those that could be harmful.

ABSTRACT. Being a parent is among the most widely shared and important identities in the world. Yet, little quantitative research is available. Here, we use NLP and machine learning to track parent identity in online forum posts to understand how a parent identity develops, affects mental health, and reacts to adversity.

ABSTRACT. Leveraging logs for Wikipedia in 12 languages, we study how COVID19 changed the way Wikipedia is read and written. Our results demonstrate the utility of Wikipedia and the resilience of its editor community in the face of adverse conditions and shed light on mechanisms behind attention and voluntary participation online.

ABSTRACT. During the early days of the pandemic, we launched an online campaign to debunk COVID-19 rumors that disseminated accurate coronavirus-related information to over 50,000 individuals. The campaign aimed to collect fact-checked information from regions that had already suffered from the infodemic and spread them to other regions where the infodemic was at its infancy. Alongside our campaign, we conducted a series of research projects to understand what kind of coronavirus-related information was being shared online. Focusing on misinformation, we quantified the spread of COVID-19 misinformation through survey studies covering more than 40 countries around the world. This talk will introduce our Facts Before Rumors campaign as well as other on-going efforts by UNICEF and WHO.

ABSTRACT. Scandals around deviant affairs, such as corruption of economic or political elites, are expected to reveal truths. In this study, we argue that public inquiries into scandals uncover only a limited amount of knowledge because of the endogenous network dynamics of the underlying denunciation process. This article analyzes the testimonies given during a Canadian Royal Commission, referred to as the Gomery Commission, which unfolded throughout 2004-2005 and inquired into the inner workings of a corruption scandal within the federal government of Canada. Analyses are based on original data of relational denunciation events that emerged from the public testimonies of 172 witnesses. Multilevel discrete-time event history analyses are applied to demonstrate the temporal and relational embeddedment of denunciation. Witnesses in the Commission’s proceedings tend to reciprocate denunciations, denounce others who have been previously denounced, and punish those who reveal too much information. We conclude that the dynamics of denunciation maintain secrecy more effectively and consistently than they produce knowledge about deviant activities.

ABSTRACT. The proliferation of social media has given rise to widespread study and speculation about the impact of digital technologies on politics, activism, and social change. Key among these debates is the role social media play in shaping the contemporary public sphere, and by proxy, our societies. Maligned by some as “slacktivism,” I will argue social media platforms such as Twitter create unique opportunities for often-excluded voices to challenge the terms of public debate. Using the evidence from Twitter hashtag networks assuch as #BlackLivesMatter and #MeToo, I will demonstrate how hashtag activism complements other forms of activism to change the terms of mainstream public debates about race and gender justice in the United States.