University of Michigan School of Information
Virtual Reality | Health Insurance | Screenshots: UMSI Research Roundup
Tuesday, 12/03/2024
By Noor HindiUniversity of Michigan School of Information faculty and PhD students are creating and sharing knowledge that helps build a better world. Here are some of their recent publications.
Publications
User-Centric Textual Descriptions of Privacy-Enhancing Technologies for Ad Tracking and Analytics
Proceedings on Privacy Enhancing Technologies Symposium, January 2025
Lu Xian, Song Mi Lee-Kan, Jane Im, Florian Schaub
Describing Privacy Enhancing Technologies (PETs) to the general public is challenging but essential to convey the privacy protections they provide. Existing research has explored the explanation of differential privacy in health contexts. Our study adapts wellperforming textual descriptions of local differential privacy from prior work to a new context and broadens the investigation to the descriptions of additional PETs. Specifically, we develop usercentric textual descriptions for popular PETs in ad tracking and analytics, including local differential privacy, federated learning with and without local differential privacy, and Google’s Topics. We examine the applicability of previous findings to these expanded contexts, and evaluate the PET descriptions with quantitative and qualitative survey data (n=306). We find that adapting a processand implications-focused approach to the ad tracking and analytics context achieved similar effects in facilitating user understanding compared to health contexts, and that our descriptions developed with this process+implications approach for the additional, understudied PETs help users understand PETs’ processes. We also find that incorporating an implications statement into PET descriptions did not hurt user comprehension but also did not achieve a significant positive effect, which contrasts prior findings in health contexts. We note that the use of technical terms as well as the machine learning aspect of PETs, even without delving into specifics, led to confusion for some respondents. Based on our findings, we offer recommendations and insights for crafting effective user-centric descriptions of privacy-enhancing technologies.
Misalignments and Demographic Differences in Expected and Actual Privacy Settings on Facebook
Proceedings on Privacy Enhancing Technologies Symposium, January 2025
Byron Lowens, Sean Scarnecchia, Jane Im, Tanisha Afnan, Annie Chen, Yixin Zou, Florian Schaub
Social media platforms pose privacy risks when data is used in unexpected ways (e.g., for advertising or data sharing with partners). Using a custom browser extension and an online survey with 195 Facebook users, we investigated (1) whether participants’ expected values of their Facebook privacy settings were (mis)aligned with their actual settings; (2) demographic differences in privacy expectation-setting mismatches; and (3) participants' privacy concerns and trust towards Facebook.Our study presents a current and comprehensive analysis of Facebook users' privacy settings. We find that expectation-setting mismatches are prevalent: all participants had at least one mismatch; many had multiple, often expecting their settings to be more restrictive than they were. We also found that Facebook's default values are not aligned with people's expectations and/or actual settings, which suggests that those defaults are ineffective. Furthermore, mismatches differed along certain demographic variables.Participants' trust in Facebook decreased after they became aware of mismatches and their actual settings. Our empirical findings indicate that, despite increased public awareness, media scrutiny, and regulatory attention regarding privacy issues, there is still a substantial and concerning disconnect between how private people perceive their social media data to be and how exposed their data actually is, opening them up to both interpersonal and institutional privacy risks. We discuss design and public policy implications of our findings.
Distinguishing social virtual reality: Comparing communication channels across perceived social affordances, privacy, and trust
Computers in Human Behavior, December 2024
James J. Cummings, Alexis Shore Ingber
Social virtual reality (SVR) attempts to allow for connections akin to face-to-face communication (Ftf). Yet, it is unclear whether the technology successfully mimics Ftf or more closely resembles other mediated communication channels. This study empirically compares SVR and other communication channels in terms of perceived social affordances, privacy, and trust through a between-subjects online survey (n = 743). Findings indicate that SVR and Ftf are similar regarding some perceived affordances (e.g., personalization) but differ with respect to others (e.g., anonymity, presence). Additionally, SVR is perceived as significantly distinct from one or multiple mediated channels for almost every measured social affordance. While SVR is seen as offering relatively greater levels of affordances that benefit interpersonal interaction, privacy concerns and a lack of trust in other users were found to often characterize the current user experience. This study provides theoretical insights for affordance research and practical implications for SVR designers.
The Future of Research on Social Technologies
CSCW Companion ‘24: Companion Publication of the 2024 Conference on Computer-Supported Cooperative Work and Social computing, November 2024
Sarita Schoenebeck, Motahhare Eslami, Eric Gilbert, Karrie Karahalios, David R Karger
While the benefits of social technologies have been profound, the harms too have cast a long shadow. This panel follows a workshop hosted by the Computing Community Consortium (CCC) Workshop “The Future of Research on Social Technologies” held in November 2023. The workshop asked two guiding questions: “What should we know about social technologies, and what is needed to get there?” This panel will discuss some key themes that arose. First, is decentralization a viable and desirable model? Second, how will AI impact online conversations? Third, how should we combat threats to researchers studying risky or contentious topics? Last, are we publishing too much while having too little impact? Panelists will present their own stances and then discuss and debate where the field should go.
Beyond “Reviewer 2” Problems: Responding to the Peer Review Crisis in Computing Research
CSCW Companion ‘24: Companion Publication of the 2024 Conference on Computer-Supported Cooperative Work and Social computing, November 2024
Jessica Vitak, Amy Bruckman, Cliff Lampe, Xinru Page, Marisol Wong-Vallacres
Peer review plays a critical role in scientific research, serving as a benchmark in assessing the quality and rigor of research papers. Publishing peer-reviewed research is also important due to its role in employment and promotion decision-making. Many fields and disciplines, however, are facing increasing challenges around peer review, and conferences and journals are struggling to find enough reviewers for the steadily increasing number of submissions and to ensure reviews are high quality. This panel features five SIGCHI members who have served in various administrative roles related to peer review. Join us for an important conversation about current challenges and potential solutions as we consider ways to better engage the CSCW community in processes essential to our continued growth and success.
Games and Play SIG at CSCW: Approaching Games Research from a CSCW Perspective
CSCW Companion ‘24: Companion Publication of the 2024 Conference on Computer-Supported Cooperative Work and Social computing, November 2024
Guo Freeman, Regan L. Mandryk, Julian Frommel, Pejman Mirza-Babaei, Cliff Lampe
The social and collaborative side of games and play has given rise to many questions about the social dynamics of digital games and their impacts on groups, teams, communities, and societies, which are highly valuable and relevant to the CSCW community. In the history of CSCW, diverse groups of researchers, students, academics, practitioners, and developers have contributed to introducing and promoting games and play research, either through studying gameplay itself or leveraging games and play as a valuable context to study other core CSCW concepts/questions. Despite the shared research interests in games and play, we understand that these groups of researchers may often lack a dedicated opportunity to get together at CSCW to connect, network, and learn from each other’s various research efforts and perspectives. We also recognize that there are currently many members of the larger CSCW community whose research and practice intersects with games and play. The main goal of this SIG is to help both games and play researchers at CSCW and CSCW researchers who are interested in knowing more about games and play research connect with each other and further build a sense of community. Together, we aim to strengthen ties between games and play research and other areas of research within CSCW and explore how games research and CSCW can benefit from each other.
Social norms: Enforcement, breakdown and polarization
European Economic Review, November 2024
Loukas Balafoutas, Eugen Dimant, Simon Gachter, Erin Krupka
The collection of scientific contributions in this special issue explores the role of social norms in guiding collective behavior and the complexities of enforcing these norms in polarized contexts. It examines how norms, while fostering social cohesion, can also contribute to societal divisions, especially in politically charged environments. The included studies highlight three key areas: the enforcement of norms through mechanisms such as information dissemination and leadership; the breakdown of norms in polarized societies, where political affiliations and trust erosion can exacerbate discrimination; and the polarization of norms across political and generational divides, which can hinder collective action and deepen societal fragmentation. Together, these contributions provide valuable insights into how norms are maintained, challenged, or eroded in diverse settings, offering guidance for strengthening social cohesion in the face of contemporary global challenges.
Psychosocial and pandemic-related circumstances of suicide deaths in 2020: Evidence from the National Violent Death Reporting System
PLOS One, October 2024
Briana Mezuk, Viktoryia Kalesnikava*, Aparna Ananthasubramaniam*, Annalise Lane, Alejandro Rodriguez-Putnam, Lily Johns, Courtney Bagge, Sarah Burgard, Kara Zivin
Purpose: To describe and explore variation in ‘pandemic-related circumstances’ among suicide decedents during the first year of the COVID-19 pandemic.
Methods: We identified pandemic-related circumstances using decedents’ text narratives in the 2020 National Violent Death Reporting System. We use time-series analysis to compare other psychosocial characteristics (e.g., mental health history, interpersonal difficulties, financial strain) of decedents pre-pandemic (2017/2018: n = 56,968 suicide and n = 7,551 undetermined deaths) to those in 2020 (n = 31,887 suicide and n = 4,100 undetermined). We characterize common themes in the narratives with pandemic-related circumstances using topic modeling, and explore variation in topics by age and other psychosocial circumstances.
Results: In 2020, n = 2,502 (6.98%) narratives described pandemic-related circumstances. Compared to other deaths in 2020 and to the pre-pandemic period, decedents with pandemic-related circumstances were older and more highly educated. Common themes of pandemic-related circumstances narratives included: concerns about shutdown restrictions, financial losses, and infection risk. Relative to decedents of the same age that did not have pandemic-related circumstances in 2020, those with pandemic-related circumstances were more likely to also have financial (e.g., for 25–44 years, 43% vs. 12%) and mental health (76% vs. 66%) psychosocial circumstances, but had similar or lower prevalence of substance abuse (47% vs. 49%) and interpersonal (40% vs. 42%) circumstances.
Conclusions: While descriptive, these findings help contextualize suicide mortality during the acute phase of the COVID-19 pandemic and can inform mental health promotion efforts during similar public health emergencies.
Supporting Driver Attention Toward Potential Hazards During Takeover: A Preliminary Result
Proceedings of the Human Factors and Ergonomics Society Annual Meeting, October 2024
Doo Won Han, Jundi Liu, X. Jessie Yang, Alicia Romo, William Horrey, Dawn Tilbury, Feng Zhou, Lionel Robert, Lisa Molnar
This study investigates the impact of a supportive system on takeover transitions in conditionally automated driving (SAE Level 3). The supportive system is designed to direct drivers’ attention toward potential hazards in the environment when a takeover request occurs. The study comprises two components: (a) identifying various types of potential hazards using naturalistic driving data, and (b) conducting a driving simulator study to develop and assess a gaze guidance system based on the N-SEEV model of visual attention. Results indicate that drivers using a highly salient attention guidance system were less likely to collide with a secondary hazard during takeover transitions. This suggests that gaze guidance support is an effective approach for assisting drivers during takeover transitions.
Client-Designer Negotiation in Data Visualization Projects
IEEE Transactions on Visualization and Computer Graphics, October 2024
Elsie Lee-Robbins, Arran Ridley, Eytan Adar
Data visualization designers and clients need to communicate effectively with each other to achieve a successful project. Unlike a personal or solo project, working with a client introduces a layer of complexity to the process. Client and designer might have different ideas about what is an acceptable solution that would satisfy the goals and constraints of the project. Thus, the client-designer relationship is an important part of the design process. To better understand the relationship, we conducted an interview study with 12 data visualization designers. We develop a model of a client-designer project space consisting of three aspects: surfacing project goals , agreeing on resource allocation , and creating a successful design . For each aspect, designer and client have their own mental model of how they envision the project. Disagreements between these models can be resolved by negotiation that brings them closer to alignment. We identified three main negotiation strategies to navigate the project space: 1) expanding the project space to consider more potential options, 2) constraining the project space to narrow in on the boundaries, and 3) shifting the project space to different options. We discuss client-designer collaboration as a negotiated relationship, with opportunities and challenges for each side. We suggest ways to mitigate challenges to avoid friction from developing into conflict.
Health Insurance Status since the Affordable Care Act among US Adults with and without CKD
Journal of the American Society of Nephrology, October 2024
Jenna Kiryakos, Tiffany Veinot, Yun Han, William Herman, Hal Morgenstern, Yoshihisa Miyamoto, Rajiv Saran, Jennifer Bragg-Gresham
Background: Adults without health insurance are more likely to delay care due to costs, including preventive care and services for acute and chronic illnesses. In 2010, the Affordable Care Act (ACA) was implemented to reduce the number of uninsured individuals under age 65 through health insurance exchanges, the dependent coverage provision, Medicaid expansion, and other policy changes. As health care coverage is important for optimal management of chronic kidney disease (CKD), we sought to examine the proportion of adults who lack health care coverage by CKD status.
Methods: We included 38,119 participants from the National Health and Nutrition Examination Survey (NHANES, 2001–March 2020) aged 18–64 years. CKD was defined by the presence of albuminuria (urine albumin to creatinine ratio – UACR ≥30 mg/g) or estimated glomerular filtration rate (eGFR) <60 ml/min/1.73m2. Lack of health care coverage was defined by a response of “No” to the NHANES question “Are you covered by health insurance or some other kind of health care plan?”. The proportion of adults without health insurance by CKD status was age-standardized to the 2010 U.S. Census population for adults ages 18–24, 25–44, and 45–64.
Results: The sample included 3,967 participants with CKD and 34,152 participants without CKD. The mean age of the entire sample was 40.7 and 50% were male. For both those with and without CKD, prevalence of lack of health insurance increased from 2001-2004 to 2009-2012 (p=0.01), then declined (p<0.001). The decline corresponded with the implementation of the ACA (Figure). Age-standardized trends were similar to crude rates.
Conclusion: Since 2009-2012, coinciding with the ACA implementation, there has been a decrease in the proportion of US adults without health insurance. This trend was present in adults with or without CKD; however, the percentage of persons lacking health insurance tended to be higher among those with CKD.
Long-Term Effects of Structural Racism on Kidney Health: Redlining in Atlanta
Journal of American Society of Nephrology, October 2024
Jennifer Bragg-Gresham, Linda Fraunhofer, Ana Laura Licon, Tiffany Veinot, Jennifer Ennis, Rajiv Saran
Background: In 1938, the Home Owners' Loan Corporation (HOLC) produced a map of Alanta, GA, as a safety investment guide for mortgage lenders. Areas labeled “hazardous” reinforced racial residential segregation with lasting patterns of continuing inequality. We hypothesize that individuals residing in “less desirable” neighborhoods continue to face challenges contributing to the risk of kidney disease.
Methods: Using Labcorp's most recent kidney disease laboratory tests (~25K, 2021-2024) from individuals residing in previously mapped HOLC districts in Atlanta, GA, we spatially joined results to each grade (A: Best [Blue], B: Desirable [Green], C: Declining [Yellow], and D: Hazardous [Red]). Kidney disease was defined by either an eGFR <60 ml/min/1.73m2 or a UACR ≥30 mg/g. Hot spot analysis was conducted employing the Getis-Ord Gi*. Logistic regression assessed the odds of kidney disease by HOLC grade, adjusted for age, sex, and current Area Deprivation Index (ADI).
Results: Individuals living in HOLC-A & B neighborhoods were older with a higher proportion of females than those in HOLC-C & D. The prevalence of CKD and current ADI were higher within each successively worse HOLC grade (CKD from 5.3%-13.2% and ADI from 1-3.6). The odds of kidney disease was significantly higher for all HOLC grades compared to grade A (B: OR=1.9, C: OR=2.9, and D: OR=3.7, all p<0.0001). Current day ADI was associated with 13% higher odds of CKD per 1 higher ADI score, p<0.0001.
Conclusion: Significant positive associations were seen between historically redlined areas of Atlanta and the odds of kidney disease. It is imperative to determine the mechanisms underlying these observations and to develop place/person-centered intervention programs to mitigate the effects of these disparities. Future work will assess the causal impact of current day social determinants of health in these geographic areas, including measures of housing, food insecurity, access to healthcare, climate/temperature, and air pollution.
Abstract 25: The Blood Pressure Effects of a Just-in-time-adaptive Intervention for Physical Activity and Diet in Patients with Hypertension: A Randomized Controlled Trial
Hypertension, October 2024
Michael Dorsch, Jessica Golbus, Tanima Basu, Evan Luff, Kimberly Warden, Michael Giacalone, Sarah Bailey, Gabriella VanAken, Sonali Mishra, Predrag Klasnja, Mark Newman, Lesli Skolarus, Brahmajee Nallamothu
Background: Emerging data suggest mobile health interventions are a promising approach for managing hypertension, but large-scale studies are lacking. The myBPmyLife mobile application is a just-in-time adaptive intervention incorporating behavioral change strategies such as goal setting, prompts, visualizations, and feedback to encourage increased physical activity and lower-sodium food choices.
Methods: The study was a prospective, randomized-controlled trial that enrolled patients with hypertension from the University of Michigan Health in Ann Arbor, MI, and the Hamilton Community Health Network, a series of federally qualified health center clinics, in Flint, MI. Participants were randomized 1:1 to either the intervention group receiving the myBPmyLife mobile application or the control group and followed for 6 months. Participants in both groups received a smartwatch and a Bluetooth blood pressure (BP) monitor and were prompted to perform BP measurements once weekly. The primary outcome was systolic BP (SBP) change from baseline to 6 months. Secondary outcomes included change from baseline to 6 months in mean daily step count assessed by smartwatch and mean daily sodium intake assessed by the Block sodium screener.
Results: The study enrolled 602 participants between December 2021 and July 2023. Mean age was 59.6 years (SD 14), with 74.4% self-identifying as White and 51.7% as men. Baseline estimated sodium intake was 3089 mg (SD 1078) per day, mean step count was 7618 (SD 3778) steps per day, SBP was 132 (SD 15) mmHg, and diastolic blood pressure (DBP) was 82 (SD 10) mmHg. Change in SBP from baseline to 6 months was -5.2 (SD 15) mmHg in the intervention and -5.7 (SD 15) mmHg in the control group (Table; p=0.76). Change in DBP from baseline to 6 months was -3.0 (SD 9) mmHg in the intervention and -3.6 (SD 10) mmHg in the control group (p=0.52). From baseline to 6 months, estimated daily sodium intake decreased by 1145 mg (SD 1023) in the intervention and 860 mg (SD 1001) in the control group (p=0.002), while mean daily step count increased by 170 steps (SD 2690) in the intervention and decreased by 319 steps (SD 2612) in the control group (p=0.040).
Conclusion: A mobile health intervention, myBPmyLife, promoting lifestyle modification in hypertensive patients did not change SBP over 6 months compared to the control group despite significantly improving daily sodium intake and step count over that period.
Feminist Interaction Techniques: Deterring Non-Consensual Screenshots with Interaction Techniques
UIST '24: Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology, October 2024
Li Qiwei, Francesca Lameiro, Shefali Patel, Cristi Isaula-Reyes, Eytan Adar, Eric Gilbert, Sarita Schoenebeck
Non-consensual Intimate Media (NCIM) refers to the distribution of sexual or intimate content without consent. NCIM is common and causes significant emotional, financial, and reputational harm. We developed Hands-Off, an interaction technique for messaging applications that deters non-consensual screenshots. Hands-Off requires recipients to perform a hand gesture in the air, above the device, to unlock media—which makes simultaneous screenshotting difficult. A lab study shows that Hands-Off gestures are easy to perform and reduce non-consensual screenshots by 67%. We conclude by generalizing this approach and introduce the idea of Feminist Interaction Techniques (FIT), interaction techniques that encode feminist values and speak to societal problems, and reflect on FIT’s opportunities and limitations.
CKD Awareness among US Adults by Kidney Disease: Improving Global Outcomes (KDIGO) Risk Classification
Journal of the American Society of Nephrology, October 2024
Jennifer Bragg Gresham, Jenna Kiryakos, Michael Heung, Tiffany Veinot, Brenda Gillespie, Hal Morgenstern, William Herman, Yoshihisa Miyamoto, Fang Xu, Rajiv Saran
Background: Nearly 1 in 7 U.S. adults have chronic kidney disease (CKD); however, 9 in 10 adults with CKD do not know of their diagnosis. Improving awareness is particularly important because newer therapies are now available that can significantly improve kidney and cardiovascular outcomes in these patients. We sought to quantify awareness of CKD by eGFR and albuminuria categories among adults with a focus on stages G1–G2 with albuminuria (A2–A3), an at-risk population for CKD progression and cardiovascular disease.
Methods: Using the National Health and Nutrition Examination Survey (2017–March 2020), CKD was based on the CKD-EPI 2021 formula and albuminuria testing (urine albumin to creatinine ratio) among U.S. adults aged ≥ 20 years (N=7,671). Awareness was defined as a self-reported affirmative response to the question “Have you ever been told by a doctor or other health professional that you had weak or failing kidneys?” Analyses accounted for complex survey design and weights.
Results: Awareness of CKD was markedly lower among adults in the earlier stages of CKD (Table), especially for those with A2 category of albuminuria with eGFR ≥ 60 ml/min/1.73m2. Extrapolating to the US population, this represents over 18 million U.S. adults with CKD who are unaware of the disease. Adults with CKD G5 had the highest awareness, yet 15-25% (approximately 52,000 adults) were unaware.
Conclusion: While earlier stages of CKD (G1–G3) are more prevalent in the U.S. population than advanced stages (G4–G5), adults with early-stage CKD are much less likely to be aware of their diagnosis. Most adults with albuminuria and eGFR ≥ 60 ml/min/1.73m2 are unaware of having kidney disease. A limitation is that this analysis uses single measurements of these kidney markers. Raising awareness of CKD in earlier stages of the disease through increase in screening among those with risk factors may help to prompt earlier implementation of preventive strategies and could be studied further.
Do I Have a Say in This, or Has ChatGPT Already Decided for Me?
XRDS: Crossroads, The ACM Magazine for Students, October 2024
It's not just about LLMs, it's about us too.
Steering UX Education: Designing an Automotive UX Course
James M. Rampton, Lionel P. Robert Jr., Myounghoon Jeon, Manhua Wang, Gayoung Ban, Ankit R. Patel, Dave B. Miller
In-car interfaces are the primary medium for communication between the occupants and the increasingly agentic vehicle systems. Although many universities teach automotive user experience and design courses, there is no consensus on what topics to cover. Some schools may choose to focus on the interior design of the cabin, including, but not limited to, physical controls and ergonomics, while other schools may just focus on the usability of what is shown to the driver and passengers. Participants in our workshop will discuss various topics for teaching Automotive UX and UI at both undergraduate and graduate levels, participating in interactive activities such as panels, breakout discussions, and syllabus design. Participants will then combine and form their findings into a course outline based on themes (ex., UI, Human Factors, etc.). This workshop is expected to achieve general consensus on a Automotive UX curriculum drawing from diverse stakeholders, including academia, industry, and government.
UI Development Experiences of Programmers with Visual Impairments in Product Teams
Equity, Diversity, and Inclusion in Software Engineering, September 2024
Maulishree Pandey, Sharvari Bondre, Vaishnav Kameswaran, Hrishikesh Rao, Sile O’Modhrain, Steve Oney
The tools and techniques that software engineers use to collaborate are critical in deciding who can contribute to software projects and the roles they can play within those teams. The consistent growth of UI developer job roles has made many programmers seek UI engineering jobs. It is important to understand the accessibility of the profession and identify ways to make it more inclusive. We conducted two qualitative studies to better understand the strategies that mixed-ability teams – specifically teams where some team members identify as having a visual impairment and some do not – use to collaborate on user interface (UI) development. In this chapter, we summarize and synthesize the findings from our prior studies to highlight the challenges programmers with visual impairments encounter in collaborative UI programming. The chapter concludes with recommendations for building more inclusive software engineering teams by fostering communication and help-seeking interactions, which we hope product teams would find valuable. We also derive implications for UI frameworks that aim to support accessible application development. These implications can inform the engineering choices of product teams as well as inform the efforts of researchers and developers building these frameworks.
Protecting Private Communications through Law and Policy: The Case of the Screenshot Feature
Communication Law and Policy, September 2024
Individuals rely on digital communication platforms to express the most intimate details of their lives with others. The screenshot feature threatens the ability to do this with confidence, authorizing nonconsensual information collection and sharing by interpersonal actors. In some instances, the harms enabled by the screenshot feature have incited legal action. This study analyzes how judges and policymakers have conceptualized and regulated screenshot collection and the sharing of private communications. Analysis of case law and Federal Trade Commission (FTC) rulemakings reveal inconsistencies both within law and policy and as compared to evidence from social-media users. Judges have provided broad definitions of “authorization” and lofty thresholds to sustain individual harm, making statutory regulation of screenshot collection and sharing unlikely. However, guidance from the FTC has demonstrated a nuanced approach to privacy that recognizes the significance of platform design. Results suggest that design-based strategies—both ex ante and ex post—would be a promising step toward adjusting the norms around screenshot collection and sharing of digital messages.
Toward a community driven approach to urban data-driven governance
International Communication Gazette, June 2024
Drawing from communication infrastructure theory, I propose a methodological and analytical intervention for the smart city and related initiatives: a community-driven approach to urban data-driven governance. This approach contends that data-related actors increasingly act as key storytelling agents within urban governance, given the rise of datafication. Moreover, it emphasizes the importance of an alternative storytelling network, positing the local storytelling network as a key—yet often minimally visible— set of experts and storytellers that can activate data-driven change and overcome extant gaps in dominant, “limited” versions of the smart city. As such, I spotlight two counterdata projects as prototypes for this intervention. In all, I argue it is imperative that urban governance explores this framework to incorporate, and design with, a wider range of local actors and knowledge in mind: not only to address critiques but, more so, to build more inclusive models in the public interest.
Data and/as activism: Community-based racial justice data repertories
New Media and Society, October 2023
This article maps the racial justice data repertoires of 11 community- and university-based projects at the nexus of racial and data justice. To examine them, I propose a typology of these co-occurring data activism strategies. Using a qualitative content analysis of organizations’ public statements (n = 74), the typology allows me to examine the diverse, strategic nature of racial justice data repertoires. Their wide array reveals the contentious racial politics of data, or the ways data reproduces racial inequalities but can also be mobilized to contest them.
Pre-prints, Working Papers, Articles, Reports, Workshops and Talks
Mapping the Podcast Ecosystem with the Structured Podcast Research Corpus
arXiv, November 2024
Benjamin Litterer, David Jurgens, Dallas Card
Podcasts provide highly diverse content to a massive listener base through a unique ondemand modality. However, limited data has prevented large-scale computational analysis of the podcast ecosystem. To fill this gap, we introduce a massive dataset of over 1.1M podcast transcripts that is largely comprehensive of all English language podcasts available through public RSS feeds from May and June of 2020. This data is not limited to text, but rather includes audio features and speaker turns for a subset of 370K episodes, and speaker role inferences and other metadata for all 1.1M episodes. Using this data, we also conduct a foundational investigation into the content, structure, and responsiveness of this ecosystem. Together, our data and analyses open the door to continued computational research of this popular and impactful medium.
Content Quality Vs. Attention Allocation: An Llm-Based Case Study In A Peer-To-Peer Mental Health Network
arXiv, November 2024
Teng Ye, Hanson Yan, Xuhuan Huang, Connor Grogan, Walter Yuan, Qiaozhu Mei, Matthew O. Jackson
With the rise of social media and peer-to-peer networks, users increasingly rely on crowdsourced responses for information and assistance. However, the mechanisms used to rank and promote responses often prioritize and end up biasing in favor of timeliness over quality, which may result in suboptimal support for help-seekers. We analyze millions of responses to mental health-related posts, utilizing large language models (LLMs) to assess the multi-dimensional quality of content, including relevance, empathy, and cultural alignment, among other aspects. Our findings reveal a mismatch between content quality and attention allocation: earlier responses—despite being relatively lower in quality—–receive disproportionately high fractions of upvotes and visibility due to platform ranking algorithms. We demonstrate that the quality of the top-ranked responses could be improved by up to 39 percent, and even the simplest re-ranking strategy could significantly improve the quality of top responses, highlighting the need for more nuanced ranking mechanisms that prioritize both timeliness and content quality, especially emotional engagement in online mental health communities.
Plurals: A System for Guiding LLMs Via Simulated Social Ensembles
arXiv, November 2024
Joshua Ashkinaze, Emily Fry, Narendra Edara, Eric Gilbert, Ceren Budak
Recent debates raised concerns that language models may favor certain viewpoints. But what if the solution is not to aim for a “view from nowhere” but rather to leverage different viewpoints? We introduce Plurals, a system and Python library for pluralistic AI deliberation. Plurals consists of Agents (LLMs, optionally with personas) which deliberate within customizable Structures, with Moderators overseeing deliberation. Plurals is a generator of simulated social ensembles. Plurals integrates with government datasets to create nationally representative personas, includes deliberation templates inspired by democratic deliberation theory, and allows users to customize both information-sharing structures and deliberation behavior within Structures. Six case studies demonstrate fidelity to theoretical constructs and efficacy. Three randomized experiments show simulated focus groups produced output resonant with an online sample of the relevant audiences (chosen over zero-shot generation in 75% of trials). Plurals is both a paradigm and a concrete system for pluralistic AI.
A Test of Time: Predicting the Sustainable Success of Online Collaboration in Wikipedia
arXiv, October 2024
Abraham Israeli, David Jurgens, Daniel Romero
The Internet has significantly expanded the potential for global collaboration, allowing millions of users to contribute to collective projects like Wikipedia. While prior work has assessed the success of online collaborations, most approaches are time-agnostic, evaluating success without considering its longevity. Research on the factors that ensure the long-term preservation of high-quality standards in online collaboration is scarce. In this study, we address this gap. We propose a novel metric, ‘Sustainable Success,’ which measures the ability of collaborative efforts to maintain their quality over time. Using Wikipedia as a case study, we introduce the SustainPedia dataset, which compiles data from over 40K Wikipedia articles, including each article’s sustainable success label and more than 300 explanatory features such as edit history, user experience, and team composition. Using this dataset, we develop machine learning models to predict the sustainable success of Wikipedia articles. Our best-performing model achieves a high AU-ROC score of 0.88 on average. Our analysis reveals important insights. For example, we find that the longer an article takes to be recognized as high-quality, the more likely it is to maintain that status over time (i.e., be sustainable). Additionally, user experience emerged as the most critical predictor of sustainability. Our analysis provides insights into broader collective actions beyond Wikipedia (e.g., online activism, crowdsourced open-source software), where the same social dynamics that drive success on Wikipedia might play a role. We make all data and code used for this study publicly available for further research.
Harnessing Causality in Reinforcement Learning With Bagged Decision Times
arXiv, October 2024
Daiqi Gao, Hsin-Yu Lai, Predrag Klasnja, Susan A. Murphy
We consider reinforcement learning (RL) for a class of problems with bagged decision times. A bag contains a finite sequence of consecutive decision times. The transition dynamics are non-Markovian and non-stationary within a bag. Further, all actions within a bag jointly impact a single reward, observed at the end of the bag. Our goal is to construct an online RL algorithm to maximize the discounted sum of the bag-specific rewards. To handle non-Markovian transitions within a bag, we utilize an expert-provided causal directed acyclic graph (DAG). Based on the DAG, we construct the states as a dynamical Bayesian sufficient statistic of the observed history, which results in Markov state transitions within and across bags. We then frame this problem as a periodic Markov decision process (MDP) that allows non-stationarity within a period. An online RL algorithm based on Bellman-equations for stationary MDPs is generalized to handle periodic MDPs. To justify the proposed RL algorithm, we show that our constructed state achieves the maximal optimal value function among all state constructions for a periodic MDP. Further we prove the Bellman optimality equations for periodic MDPs. We evaluate the proposed method on testbed variants, constructed with real data from a mobile health clinical trial.
SPRIG: Improving Large Language Model Performance by System Prompt Optimization
arXiv, October 2024
Lechen Zhang, Tolga Ergen, Lajanugen Logeswaran, Moontae Lee, David Jurgens
Large Language Models (LLMs) have shown impressive capabilities in many scenarios, but their performance depends, in part, on the choice of prompt. Past research has focused on optimizing prompts specific to a task. However, much less attention has been given to optimizing the general instructions included in a prompt, known as a system prompt. To address this gap, we propose SPRIG, an editbased genetic algorithm that iteratively constructs prompts from prespecified components to maximize the model’s performance in general scenarios. We evaluate the performance of system prompts on a collection of 47 different types of tasks to ensure generalizability. Our study finds that a single optimized system prompt performs on par with task prompts optimized for each individual task. Moreover, combining system and task-level optimizations leads to further improvement, which showcases their complementary nature. Experiments also reveal that the optimized system prompts generalize effectively across model families, parameter sizes, and languages. This study provides insights into the role of system-level instructions in maximizing LLM potential.
Making Exploratory Search Engines using Qualitative Case studies: a mixed method implementation using interviews with Detroit Artisans
ResearchGate, October 2024
Kwame Porter Robinson, Matthew A. Garvin, Ron Eglash, Lionel Peter Robert
Search engine algorithms are increasingly subjects of critique, with evidence indicating their role in driving polarization, exclusion, and algorithmic social harms. Many proposed solutions take a top-down approach, with experts proposing bias-corrections. A more participatory approach may be possible, with those made vulnerable by algorithmic unfairness having a voice in how they want to be "found." By using a mixed methods approach, we sought to develop search engine criteria from the bottom-up. In this project we worked with a group of 16 African American artisanal entrepreneurs in Detroit Michigan, with a majority female and all from low-income communities. Through regular in-depth interviews with select participants, they highlighted their important services, identities and practices. We then used causal set relations with natural language processing to match queries with their qualitative narratives. We refer to this two-step process-deliberately focusing on social groups with unaddressed needs, and carefully translating narratives to computationally accessible forms-as a "content aware" approach. The resulting content aware search outcomes place themes that participants value, in particular greater relationality, much earlier in the list of results when compared with a standard Web search. More broadly, our use of participatory design with "content awareness" adds evidence to the importance of addressing algorithmic bias by considering who gets to address it; and, that participatory search engine criteria can be modeled as robust linkages between interviews and semantic similarity using causal set relations.
The West African X and the Kongo Cross: a Geometric Distinction for African Heritage Designs in the Americas
ResearchGate, September 2024
The Kongo cosmogram--typically portrayed as the axial alignment of a “+” sign--has been a central focus for the interpretation of visual patterns created by enslaved Africans and their descendents in the Americas. But many of these patterns are not a crossing of vertical and horizontal lines. They are, rather, diagonal lines. This makes sense in light of two facts: first, the spiritual significance of the X pattern in West African designs. Second, the fact that enslaved peoples from these West African areas were often present in the geographic areas of the Americas where the X pattern is prevalent in historical artifacts. By distinguishing between the axial alignment of the Kongo cross, and the diagonal alignment of the West African X, we can gain new insight into African heritage on both sides of the Atlantic.
Real or Robotic? Assessing Whether LLMs Accurately Simulate Qualities of Human Responses in Dialogue
arXiv, September 2024
Jonathan Ivey*, Shivani Kumar*, Jiayu Liu*, Hua Shen*, Sushrita Rakshit*, Rohan Raju*, Haotian Zhang*, Aparna Ananthasubramaniam*, Junghwan Kim*, Bowen Yi*, Dustin Wright*, Abraham Israeli*, Anders Giovanni Møller*, Lechen Zhang*, David Jurgens
Studying and building datasets for dialogue tasks is both expensive and time-consuming due to the need to recruit, train, and collect data from study participants. In response, much recent work has sought to use large language models (LLMs) to simulate both human-human and human-LLM interactions, as they have been shown to generate convincingly humanlike text in many settings. However, to what extent do LLM-based simulations actually reflect human dialogues? In this work, we answer this question by generating a large-scale dataset of 100,000 paired LLM-LLM and human-LLM dialogues from the WildChat dataset and quantifying how well the LLM simulations align with their human counterparts. Overall, we find relatively low alignment between simulations and human interactions, demonstrating a systematic divergence along the multiple textual properties, including style and content. Further, in comparisons of English, Chinese, and Russian dialogues, we find that models perform similarly. Our results suggest that LLMs generally perform better when the human themself writes in a way that is more similar to the LLM’s own style.
Toward Satisfactory Public Accessibility: A Crowdsourcing Approach Through Online Reviews to Inclusive Urban Design
arXiv, September 2024
Lingyao Li, Songhua Hu, Yinpei Dai, Min Deng, Parisa Momeni, Gabriel Laverghetta, Lizhou Fan, Zihui Ma, Xi Wang, Siyuan Ma, Jay Ligatti, Libby Hemphill
As urban populations grow, the need for accessible urban design has become urgent. Traditional survey methods for assessing public perceptions of accessibility are often limited in scope. Crowdsourcing via online reviews offers a valuable alternative to understanding public perceptions, and advancements in large language models can facilitate their use. This study uses Google Maps reviews across the United States and fine-tunes Llama 3 model with the Low-Rank Adaptation technique to analyze public sentiment on accessibility. At the POI level, most categories—restaurants, retail, hotels, and healthcare—show negative sentiments. Socio-spatial analysis reveals that areas with higher proportions of white residents and greater socioeconomic status report more positive sentiment, while areas with more elderly, highly-educated residents exhibit more negative sentiment. Interestingly, no clear link is found between the presence of disabilities and public sentiments. Overall, this study highlights the potential of crowdsourcing for identifying accessibility challenges and providing insights for urban planners.
Communication Strategies for Improving Performance in Virtual Teams: Lessons from Dota 2
TechRxiv, September 2024
Ji Eun Kim, Lingyao Li, Libby Hemphill
This study explores the impact of communication factors on team performance within the context of Multiplayer Online Battle Arena (MOBA) games by analyzing a sample of 30,473 ad hoc virtual teams in the popular game, Dota 2. We focus on players' use of computer-mediated communication tools (pings and text chat) as well as other communication-related features, such as toxicity and conversational resilience. Using logistic regression models, we examine how these factors influence team performance, specifically winning the game. Our results indicate that teams with one or more pre-existing parties (i.e., groups of players who know each other) outperform teams of strangers under low and medium communication frequency conditions. However, active communication helps overcome the disadvantages of playing with randomly assigned teammates. We also found that toxicity is negatively associated with team performance and has a greater impact on less-skilled teams. Lastly, teams that recover faster from toxic communication perform better. These results confirm the importance of communication for improving team performance and suggest tailored communication strategies for different types of teams in the MOBA game setting.
GenAI Advertising: Risks of Personalizing Ads with LLMs
arXiv, September 2024
Brian Jay Tang, Kaiwen Sun, Noah T. Curran, Florian Schaub, Kang G. Shin
Recent advances in large language models have enabled the creation of highly effective chatbots, which may serve as a platform for targeted advertising. This paper investigates the risks of personalizing advertising in chatbots to their users. We developed a chatbot that embeds personalized product advertisements within LLM responses, inspired by similar forays by AI companies. Our benchmarks show that ad injection impacted certain LLM attribute performance, particularly response desirability. We conducted a between-subjects experiment with 179 participants using chabots with no ads, unlabeled targeted ads, and labeled targeted ads. Results revealed that participants struggled to detect chatbot ads and unlabeled advertising chatbot responses were rated higher. Yet, once disclosed, participants found the use of ads embedded in LLM responses to be manipulative, less trustworthy, and intrusive. Participants tried changing their privacy settings via chat interface rather than the disclosure. Our findings highlight ethical issues with integrating advertising into chatbot responses.
RELATED
Keep up with research from UMSI experts by subscribing to our free research roundup newsletter!
