When: October 4, 2023, 12:00pm-1:00pm EDT

Where: Luddy Center for Artificial Intelligence 2005

Register: https://iu.zoom.us/meeting/register/tZIldu6spzojEte0cTVN4D1ZD_pZzmBDeBAv

Speaker: Luca Luceri

Title: AI-Driven Approaches for Countering Influence Campaigns in Socio-Technical Systems

Abstract:

The proliferation of online platforms and social media has sparked a surge in information operations designed to manipulate public opinion on a massive scale, posing significant harm at both the individual and societal levels. In this talk, I will outline a research agenda focused on identifying, investigating, and mitigating orchestrated influence campaigns and deceptive activities within socio-technical systems. I will start by detailing my research efforts in designing AI-based approaches for detecting state-backed troll accounts on social media. Modeling human decision-making as a Markov Decision Process and using an Inverse Reinforcement Learning framework, I will illustrate how we can extract the incentives that social media users respond to and differentiate genuine users from state-sponsored operators. Next, I will delve into a set of innovative approaches I developed to uncover signals of inauthentic, coordinated behaviors. By combining embedding techniques to unveil unexpected similarities in the activity patterns of social media users, along with graph decomposition methods, I will show how we can reveal network structures that pinpoint coordinated groups orchestrating information operations. Through these approaches, I will provide actionable insights to inform regulators in shaping strategies to tame harm, discussing challenges and opportunities to improve the resilience of the information ecosystem, including the potential for interdisciplinary collaborations to address these complex issues.

This is a hybrid event, hosted by Observatory on Social Media and co-hosted by the Center for Complex Networks and Systems Research. The talk will be held in room 2005 of the Luddy AI building. Please indicate if you will be attending in-person as food will be provided.

Biography: Luca Luceri is a Research Scientist at the Information Sciences Institute (ISI) at the University of Southern California (USC). His research incorporates machine learning, data and network science, with a primary focus on detecting and mitigating online harms in socio-technical systems. He investigates deceptive and malicious behaviors on social media, with a particular emphasis on problems such as social media manipulation, (mis-)information campaigns, and Internet-mediated radicalization processes. His research advances AI/ML/NLP for social good, computational social science, and human-machine interaction. In his role as a Research Scientist at ISI, Luca Luceri serves as a co-PI of the DARPA-funded program INCAS, aiming to develop techniques to detect, characterize, and track geopolitical influence campaigns. Additionally, he is the co-PI of a Swiss NSF-sponsored project called CARISMA, which develops network models to simulate the effects of moderation policies to combat online harms.

When: September 11, 2023, 1:00pm EDT

Where: Luddy Center for Artificial Intelligence 2005

Zoom link: https://iu.zoom.us/webinar/89107137699

Speaker: Marios Papachristou

Title: Resource Allocation in a Financial Contagion Environment

Abstract: 

The pandemic has spread uncertainty among financial entities that experience income shocks. A policy framework is stimulus checks, i.e. cash injections into the financial system so that consumption is stimulated, and contagion is averted. A cardinal question faced by policymakers is: Who gets the subsidies? A common pattern in these scenarios is that when somebody’s income is below a certain threshold or satisfies some criterion then the household is entitled to a stimulus check of fixed value which can depend on the entity’s features. However, such rules are limited by ignoring contagion effects through the financial network. If a business defaults that may translate to job loss for the employees who in their turn may not be able to pay their own debts, potentially creating a sequence of defaults. Our work studies resource allocations under financial contagion, through the lens of approximation algorithms and fairness. In closing, we test our methods with real-world and semi-artificial data and compare them to heuristics.
In the sequel, we generalize the static framework to the dynamic setting in which liabilities that are not paid at a certain round accumulate to the next round, as an MDP. We study fractional and (approximate) discrete bailout allocation scenarios. Our framework is vastly applicable to a variety of domains: Specifically, in any problem that corresponds to a supply and demand network that evolves over time for which the nodes that cannot meet their demand have to split it proportionally and the planner wants to allocate resources can be captured by this framework. Applications beyond financial transaction networks include ridesharing, high-performance computing, ad placement, influence maximization, financial transaction networks on the Web, etc. Finally, we extend the notions of fairness to the dynamic setting. This talk includes joint work with Jon Kleinberg and Sid Banerjee.

Biography: Marios is a 4th-year Ph.D. Candidate in the Computer Science Department at Cornell University, advised by Prof. Jon Kleinberg. His interests span theoretical and practical aspects of information networks. So far, his work has evolved around statistical network models, hypergraphs, and financial contagion. Marios has also worked at Twitter Cortex Applied Research, where he did research in scalable graph machine learning methods, and at the Office of Applied Research at Microsoft, where he did research on the intersection of large language models and multi-agent systems. His research is supported by a scholarship from the Onassis Foundation and has been supported in the past by a LinkedIn PhD Fellowship, a Cornell Fellowship, and grants from the A.G. Leventis Foundation and the Gerondelis Foundation.

When: September 15, 2023, 11:30am EDT

Where: Luddy Center for Artificial Intelligence 2005 (remote talk-feel free to attend with others)

Zoom link: https://iu.zoom.us/webinar/89107137699

Speaker: Yu Tian

Title: Structural Balance and Random Walks on Complex Networks with Complex Weights

Abstract: Complex numbers define the relationship between entities in many situations. A canonical example would be the off-diagonal terms in a Hamiltonian matrix in quantum physics. Recent years have seen an increasing interest to extend the tools of network science when the weight of edges are complex numbers. Here, we focus on the case when the weight matrix is Hermitian, a reasonable assumption in many applications, and investigate both structural and dynamical properties of the complex-weighted networks. Building on concepts from signed graphs, we  introduce a classification of complex-weighted networks based on the notion of structural balance, and  illustrate the shared spectral properties within each type. We then apply the results to characterise the  dynamics of random walks on complex-weighted networks, where local consensus can be achieved asymptotically when the graph is structurally balanced, while global consensus will be obtained when it is strictly unbalanced. Finally, we explore potential applications of our findings by generalising the notion of cut, and propose an associated spectral clustering algorithm. We also provide further characteristics of the magnetic Laplacian, associating directed networks to complex-weighted ones. The performance of the algorithm is verified on both synthetic and real networks.

Biography:

Yu Tian is a research fellow at Nordita, funded by Wallenberg Initiative on Networks and Quantum Information. Yu’s research involves various aspects of network science, including dynamics and optimisation, incorporating negative signs (e.g., friend and foe relationship) and complex weights, and community detection. Before this, Yu received her PhD from University of Oxford, where she was supervised by Prof. Renaud Lambiotte and also in collaboration with Tesco Data Science Team.

When: June 23, 2023 2pm EDT

Where: Luddy Center for Artificial Intelligence 2005 (in-person)

Speaker: Diego R. Amancio

Title: Examining the Influence of top collaborators on authors’ metrics

Abstract:

Science has become more collaborative over the past years. Although various aspects of scientific collaboration have been investigated, our study specifically focused on examining the influence of the most significant collaborator on researcher metrics across diverse disciplines. Our analysis showed that the impact of the top collaborator is dependent upon the specific field of study and even highly cited authors may co-author a substantial number of papers and accumulate a considerable proportion of citations with their most important collaborator.  These findings emphasize the significance of top collaborators in shaping research outcomes.

Biography:

Diego R. Amancio is a faculty member at the University of São Paulo, Brazil. His interests include Science of Science, Network Science, Natural Language Processing and Machine Learning.

When: April 03, 2023 12pm EST

Where: Luddy Center for Artificial Intelligence (LU2005)

Speaker: Ankur Mani

Title: Investigating Churn in Online Wellness Programs: Evidence from a U.S. Online Social Network

Abstract:

Online wellness activity platforms increasingly utilize wellness programs and social support to motivate healthy activities and improve user engagement. However, many wellness programs suffer from high churn rates that discount their expected efficacy, and negative social influence may lead to a churn contagion that amplifies the churn speed and scale. Hence, a need arises to understand why users churn wellness programs and how social contagion contributes to the churns. Leveraging the exercise challenge setting, the exercise data, and a large social network on a renowned U.S. online fitness platform, we investigate the effect of peers’ behavior in exercise challenge churn on ego. To achieve the research goal, we employ an instrumental variable framework (shown in the figure below), using the exogenous variation of peers’ weather in locations that differ from the ego’s location as instruments. The framework untangles the endogeneity of the estimated effect using variations created by peers’ weather as a shock to the ego’s churn. We measure churn as a decision an ego makes after being inactive for one to two weeks and define peers as the ones an ego follows on the platform. We find that exercise challenge churn is socially contagious and demonstrates a complex contagion. Interestingly, our analyses reveal that the social contagion of churn diffuses from the sub-central or peripheral egos with fewer friends in the social network to central egos with more friends in the social network. Such churn contagion is mostly confined to low-density network communities with members who are poorly connected. Our findings have important implications for designing intervention plans to stop online wellness program churn based on social contagion.

This is joint work with my student Yi Zhu.

Biography:

Ankur Mani is an assistant professor in the Industrial and Systems Engineering department at the University of Minnesota. He is also an affiliate of the Data Science Initiative and Control Systems and Dynamics Group at the University of Minnesota. He received his Ph.D. from the Massachusetts Institute of Technology and B.Tech. degree from the Indian Institute of Technology, Delhi. Ankur’s research interests include networks, distributed experimentation, and game theory with applications in social networks, supply chain networks, transportation networks and health care. His research has been published in major journals including Management Science and Nature Human Behavior among others and has received recognitions from the INFORMS revenue management section, INFORMS aviation section, and POMS, among others.

When: November 11, 2022 12pm EST

Where: Remote Talk

Zoom link: https://iu.zoom.us/j/89873992894?pwd=QmxmaXZaakNFSUpnQm1jT1I3T1ZOZz09

Recording: Jonas Juul Cascades Talk

Speaker: Jonas Juul

Title: Harder, better, faster, stronger cascades — or simply larger?

Abstract:

Do some types of online content spread faster or further than others? In recent years, many studies have sought answers to such questions by comparing statistical properties of network paths taken by different kinds of content diffusing online. Here we demonstrate the importance of controlling for correlations in the statistical properties being compared. In particular, we show that previously reported structural differences between diffusion paths of false and true news on Twitter disappear when comparing only cascades of the same size; differences between diffusion paths of images, videos, news, and petitions persist. Paired with a theoretical analysis of diffusion processes, our results suggest that in order to limit the spread of false news it is enough to focus on reducing the mean “infectiousness” of the information.
Joint work with Johan Ugander (Stanford University)

Biography:

Jonas L. Juul is a Carlsberg Fellow postdoc at the Technical University of Denmark. His research focuses on spreading processes and networks. His recent interests include developing methods to infer how content spreads online from observed diffusion paths, and evaluating the efficiency of mitigation measures in epidemiology. Before joining the Technical University of Denmark he was a post-doctoral researcher at the Center for Applied Mathematics, Cornell University where he worked with Austin Benson, Jon Kleinberg and Steven Strogatz. He obtained his Ph.D. in Physics of Complex Systems from the Niels Bohr Institute in 2020.

When: January 19, 2022,  2:00pm

Where: Luddy Center for Artificial Intelligence (2044)

Zoom link: https://iu.zoom.us/j/89107137699

Speaker: Brett Buttliere

Abstract:

The talk will focus on the psychology of scientist in both doing science and diffusing science, focusing in particular on the role of cognitive conflict as a motivating factor for communicating and doing science, online. Studies will be presented suggesting that people are more motivated to respond to those they disagree with, and that they write longer and more negative responses, but also that these posts are considered of higher informational content and quality. These posts also receive the most attention, and is distinctly different then situations where users do not have the opportunity to respond, where users prefer to avoid information they disagree with (filter bubble). The affordances of the situation appear to solve this large contradiction in the literature between filter bubbles and negativity biases online. Within science, studies will be outlined where scientists write more papers about negative topics, and that negative topics are discussed more online. I will also briefly present ongoing work examining the Wikipedia profiles of the world’s scientists, and open a discussion about what can and should be learned from these data (e.g., tracing the growth and movement of various fields across space and time). I hope it will be of interest to you and that there will be much room for discussion after.

Biography:

When: October 15, 2021,  12:00pm

Where: Luddy 1106 (Dorsey Learning Hall)

Zoom link: https://iu.zoom.us/j/89107137699

Speaker: Francesco Pierri 

Title: Characterization and detection of disinformation spreading in online social networks

Abstract:

Online social media expose us to a variety of false and misleading information which erodes public trust towards institutions, with severe backlashes in the real world. One example is the on-going COVID-19 pandemic, as the world experiences an “infodemic”, an overabundance of information including false and misleading content, which undermines medical intervention to fight the disease.

In this talk, I will present results from the research carried out during the last three years as a Ph.D. candidate in Politecnico di Milano, and that will be part of my forthcoming Ph.D. dissertation. I will also present results from my on-going collaboration with the Observatory on Social Media.

I leveraged a computer science and network science approach to tackle the problem of disinformation spreading in online social networks from two perspectives: (1) characterization, i.e., understanding the mechanisms and the actors involved in the spread of false and misleading information on online social media during relevant events such as political elections and the on-going COVID-19 pandemic; (2) detection, i.e., building a methodology to accurately classify news articles based on the interactions between users that take place on platforms like Twitter.

Biography:

I am a 3^rd (last) year Ph.D. student in the “Data Analytics and Decision Sciences” Ph.D. program at Politecnico di Milano, under the supervision of prof. Stefano Ceri and prof. Fabio Pammolli.

The focus of my research (and my Ph.D. dissertation) is to understand the spread of disinformation in online social networks, with both a computer science and network science approach. I also recently investigated the socio-economic consequences of the COVID-19 pandemic by leveraging human mobility data from mobile phones.

I am currently a visiting scholar at the Observatory on Social Media (since September 2020), where we investigate the COVID-19 infodemic and the spread of vaccine-related misinformation.

When: Friday, May 14, 2021

Where: https://iu.zoom.us/j/87346711649

Speaker: Chris Connell

Title: The DynACPD network embedding algorithm for prediction tasks on dynamic networks

Abstract: Classical network embeddings create a low dimensional representation of the learned relationships between features across nodes. Such embeddings are important for tasks such as link prediction and node classification. We consider low dimensional embeddings of “dynamic networks” — a family of time varying networks where there exist both temporal and spatial link relationships between nodes. We present novel embedding methods for a dynamic network based on higher order tensor decompositions for tensorial representations of the dynamic network. Our embeddings are analogous to certain classical spectral embedding methods for static networks. We demonstrate the effectiveness of our approach by comparing our algorithms’ performance on the link prediction task against an array of current baseline methods across three distinct real-world dynamic networks. Finally, we provide a mathematical rationale for this effectiveness in the regime of small incremental changes. This is joint work with Yang Wang.

Biography:

Chris Connell is interested in problems at the interface between dynamical systems, random walks, and the topology and geometry surrounding nonpositive curvature. Some of his recent work emphasizes bringing tools from these disciplines to bear on network embedding problems. He received his PhD from the University of Michigan and is now a professor of mathematics at Indiana University Bloomington.

When: Wednesday, October 2, 2019, 2pm

Where: Informatics East, room 322

Speaker: Jean-Gabriel Young

Efficient and fully Bayesian inference of complex networks from noisy data

Abstract: Rarely do we have access to error-free measurements of networks. Instead, we typically get to observe sequences of states that are, at best, indirect observations of a system’s structure. Recent research has led to a formalization of just how much, and in which ways, these measurements can inform us on the structure of real complex networks. It is now understood that domain-agnostic models are not silver bullets—one can come up with many different models of how a data set maps to a network, all leading to different inferences. The motivation for the work presented here is the realization that this leads to a tension preventing a broad adoption of these network reconstruction methods by practitioners.  On the one hand, domain expertise must necessarily go into devising good models, as to avoid erroneous conclusions. But at the same time, designing models can be challenging because one has to: derive a complete inference procedure from scratch; implement this procedure; verify the inference; and start anew if the results are not correct—for every model.

In this presentation, I will introduce a Bayesian framework that abstracts away most of the computational work, putting flexible model design center stage. The crucial modeling task that our framework leaves to the practitioner is that of determining how individual pair-wise measurements of interaction are explained by the presence or absence of an edge between two nodes. The method is broadly applicable in that these measurements can be anything, from a straightforward number of observed interactions, to time-series, or pairs comprising of a number of attempted and successful observations.

Biography:

Jean-Gabriel Young is broadly interested in problems at the intersection of statistics and complex systems. His recent work focuses on new exciting inference problems in network science, including the inference of the past of dynamical networks, network reconstruction from noisy data; and the inference of high-order interactions from pairwise data. He received his Ph.D. in Physics from Université Laval (2018), and is now a James S. McDonnell postdoctoral Fellow in complexity at the University of Michigan.