University 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 

Constructing risk in trustworthy digital repositories

Journal of Documentation, August 2024

Rebecca D. Frank

Purpose: This article investigates the construction of risk within trustworthy digital repository audits. It contends that risk is a social construct, and social factors influence how stakeholders in digital preservation processes comprehend and react to risk.

Design/methodology/approach: This research employs a qualitative research design involving in-depth semi-structured interviews with stakeholders in the Trustworthy Digital Repository Audit and Certification (TRAC) process, and document analysis of the TRAC checklist and audit reports. I apply an analytic framework based on the Model for the Social Construction of Risk in Digital Preservation to this data.

Findings: The findings validate the argument that risk in digital preservation is indeed socially constructed and demonstrate that the eight factors in the Model for the Social Construction of Risk in Digital Preservation do indeed influence how stakeholders constructed their understanding of risk. Of the eight factors in the model, communication, expertise, uncertainty and vulnerability were found to be the most influential in the construction of risk during the TRAC audit process. The influence of complexity, organizations political culture, were more limited.

Originality/value: This article brings new insights to digital preservation by demonstrating the importance of understanding risk as a social construct. I argue that risk identification and/or assessment is only the first step in the long-term preservation of digital information and show that perceptions of risk in digital preservation are shaped by social factors by applying theories of social construction and risk perception to an analysis of the TRAC process.

Transformative provenance: memory work in the Palestinian diaspora

Archival Science, September 2024

Tam Rayan

Provenance, as the foundational principle of archival studies, dictates that records with the same creator should be organized separately from those of a different creator. This idea of provenance, however, fails to consider different epistemologies of origin. Using an ethnographic study of the Palestinian diaspora in Southeast Michigan, this paper interrogates provenance through Palestinian epistemology and Palestinian futurism to theorize a transformative provenance that positions archival origins as both spatially and temporally unfixed. Rather than rejecting provenance, the concept is a useful departure point to consider how Western understandings of origin and custody can be broadened by other ways of knowing. In this article, I track the origins and custody of memories and stories, the main medium of records in this community, to highlight the culturally specific epistemologies involved in their preservation. I then propose a transformative provenance based on three qualities. First, intergenerational: Chain of custody belongs to both the past and the present, as stories belong to the time of a grandparent’s past exile and a grandchild’s present diaspora. Second, collective: With the spatial referent of memories being lost, ownership is shared within village kinship networks. Third, imaginative: Origins of memories exist in the past, present, and the future, as those in diaspora use memories to imagine future liberation. By grounding this analysis of Palestinian memory work within the community’s conceptualizations of knowledge organization, this paper contributes to current discourse around decolonial recordkeeping, non-Western epistemology, and the management of diaspora archives.

Building Contextualized Trust Profiles in Conditionally Automated Driving

IEEE Transactions on Human-Machine Systems, September 2024

Lilit Avetisyan, Jackie Ayoub, X. Jessie Yang

Trust is crucial for ensuring the safety, security, and widespread adoption of automated vehicles (AVs), and if trust is lacking, drivers and the general public may hesitate to embrace this technology. This research seeks to investigate contextualized trust profiles in order to create personalized experiences for drivers in AVs with varying levels of reliability. A driving simulator experiment involving 70 participants revealed three distinct contextualized trust profiles (i.e., confident copilots, myopic pragmatists, and reluctant automators) identified through K-means clustering, and analyzed in relation to drivers’ dynamic trust, dispositional trust, initial learned trust, personality traits, and emotions. The experiment encompassed eight scenarios where participants were requested to take over control from the AV in three conditions: a control condition, a false alarm condition, and a miss condition. To validate the models, a multinomial logistic regression model was constructed using the shapley additive explanations explainer to determine the most influential features in predicting contextualized trust profiles, achieving an F1-score of 0.90 and an accuracy of 0.89. In addition, an examination of how individual factors impact contextualized trust profiles provided valuable insights into trust dynamics from a user-centric perspective. The outcomes of this research hold significant implications for the development of personalized in-vehicle trust monitoring and calibration systems to modulate drivers’ trust levels, thereby enhancing safety and user experience in automated driving. 

Understanding the rationales and information environments for early, late, and nonadopters of the COVID-19 vaccine

Npj Vaccines, September 2024 

Lisa Singh, Le Bao, Leticia Bode, Ceren Budak, Josh Pasek, Trivellore Raghunathan, Michael Traugott, Yanchen Wang, Nathan Wycoff

Anti-vaccine sentiment during the COVID-19 pandemic grew at an alarming rate, leaving much to understand about the relationship between people’s vaccination status and the information they were exposed to. This study investigated the relationship between vaccine behavior, decision rationales, and information exposure on social media over time. Using a cohort study that consisted of a nationally representative survey of American adults, three subpopulations (early adopters, late adopters, and nonadopters) were analyzed through a combination of statistical analysis, network analysis, and semi-supervised topic modeling. The main reasons Americans reported choosing to get vaccinated were safety and health. However, work requirements and travel were more important for late adopters than early adopters (95% CI on OR of [0.121, 0.453]). While late adopters’ and nonadopters’ primary reason for not getting vaccinated was it being too early, late adopters also mentioned safety issues more often and nonadopters mentioned government distrust (95% CI on OR of [0.125, 0.763]). Among those who shared Twitter/X accounts, early adopters and nonadopters followed a larger fraction of highly partisan political accounts compared to late adopters, and late adopters were exposed to more neutral and pro-vaccine messaging than nonadopters. Together, these findings suggest that the decision-making process and the information environments of these subpopulations have notable differences, and any online vaccination campaigns need to consider these differences when attempting to provide accurate vaccine information to all three subpopulations.

Do policy instruments that restrict social identity expression increase economic cooperation? 

European Economic Review, November 2024

Daphne Chang, Roy Chen, Erin Krupka, Zhewei Song

Recent public and corporate policies restricting social identity expression, such as the face-covering restrictions in many European countries, presume that prominent signals of our social identity differences drive division even when inference about social identity is unaffected. Social identity theory predicts that limiting identity expression could positively or negatively affect how groups interact. We use an experiment to test whether a treatment that bans displaying an identity pin affects cooperation in public goods provision. Our subjects are U.K. residents who were in favor of leaving or remaining in the European Union. Each subject is simultaneously in two different yet economically identical environments that are distinguished only by the social identities of the group members. They play two simultaneous one-shot public goods games, one with others who share their identity (in-group public good), and one with a mixture of Leavers and Remainers (mixed-group public good). The political identities of all subjects and the structure of each group are known by everyone. Our treatments vary whether there exists a ban on displaying a Leaver/Remainer identity pin to others and whether Leavers or Remainers are the majority identity in the mixed groups. We find that banning pinning increases contributions to the mixed group when Leavers are the majority. These increases can be explained by changes in beliefs rather than the notion that shared group identity per se affects behavior. These setting- and identity-specific results suggest that policies designed to promote integration should be examined in the context in which they will be applied.

ScrapeViz: Hierarchical Representations for Web Scraping Macros

2024 IEEE Symposium on Visual Languages and Human-Centric Computing, September 2024

Rebecca Krosnick, Steve Oney

Programming-by-demonstration (PBD) makes it possible to create web scraping macros without writing code. However, it can still be challenging for users to understand the exact scraping behavior that is inferred and to verify that the scraped data is correct, especially when scraping occurs across multiple pages. We present ScrapeViz, a new PBD tool for authoring and visualizing hierarchical web scraping macros. ScrapeViz’s key novelty is in providing a visual representation of web scraping macros—the sequences of pages visited, generalized scraping behavior across similar pages, and data provenance. We conducted a lab study with 12 participants comparing ScrapeViz to the existing web scraping tool Rousillon and saw that participants found ScrapeViz helpful for understanding high-level scraping behavior, tracing the source of scraped data, identifying anomalies, and validating macros while authoring.

Changing hearts and minds: theorizing how, when, and under what conditions three social influence implementation strategies work

Frontiers in Health Services, September 2024

Bryan J. Weiner, Rosemary D. Meza, Predrag Klasnja, Rebecca Lengnick-Hall, Gretchen J. Buchanan, Aaron R. Lyon, Kayne D. Mettert, Marcella H. Boynton, Byron J. Powell, Cara C. Lewis

Background: Opinion leadership, educational outreach visiting, and innovation championing are commonly used strategies to address barriers to implementing innovations and evidence-based practices in healthcare settings. Despite voluminous research, ambiguities persist in how these strategies work and under what conditions they work well, work poorly, or work at all. The current paper develops middle-range theories to address this gap. 

Methods: Conceptual articles, systematic reviews, and empirical studies informed the development of causal pathway diagrams (CPDs). CPDs are visualization tools for depicting and theorizing about the causal process through which strategies operate, including the mechanisms they activate, the barriers they address, and the proximal and distal outcomes they produce. CPDs also clarify the contextual conditions (i.e., preconditions and moderators) that influence whether, and to what extent, the strategy's causal process unfolds successfully. Expert panels of implementation scientists and health professionals rated the plausibility of these preliminary CPDs and offered comments and suggestions on them. 

Findings: Theoretically, opinion leadership addresses potential adopters' uncertainty about likely consequences of innovation use (determinant) by promoting positive attitude formation about the innovation (mechanism), which results in an adoption decision (proximal outcome), which leads to innovation use (intermediate outcome). As this causal process repeats, penetration, or spread of innovation use, occurs (distal outcome). Educational outreach visiting addresses knowledge barriers, attitudinal barriers, and behavioral barriers (determinants) by promoting critical thinking and reflection about evidence and practice (mechanism), which results in behavioral intention (proximal outcome), behavior change (intermediate outcome), and fidelity, or guideline adherence (distal outcome). Innovation championing addresses organizational inertia, indifference, and resistance (determinants) by promoting buy-in to the vision, fostering a positive implementation climate, and increasing collective efficacy (mechanisms), which leads to participation in implementation activities (proximal outcome), initial use of the innovation with increasing skill (intermediate outcome) and, ultimately, greater penetration and fidelity (distal outcomes). Experts found the preliminary CPDs plausible or highly plausible and suggested additional mechanisms, moderators, and preconditions, which were used to amend the initial CPD. 

Discussion: The middle-range theories depicted in the CPDs furnish testable propositions for implementation research and offer guidance for selecting, designing, and evaluating these social influence implementation strategies in both research studies and practice settings.

The First Workshop on AI Behavioral Science

KDD '24: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, August 2024

Himabindu Lakkaraju, Qiaozhu Mei, Chenhao Tan, Jie Tang, Yutong Xie

This workshop initiates a new study field which may be named AI behavioral science. It discusses recent findings, methodologies, applications, and potential societal impacts that are related to analyzing, understanding, and directing the behaviors of AI models, especially those built upon large language models. This half-day workshop includes several keynote and invited talks, a poster session, and a panel discussion.

Surveying the Use of Social Media Data and Natural Language Processing Techniques to Investigate Natural Disasters

Natural Hazards Review, August 2024

Zihui Ma, Lingyao Li, Yujie Mao, Yu Wang, Olivia Grace Patsy, Michelle T. Bensi, Libby Hemphill, Gregory B. Baecher

The use of social media in crisis informatics has become increasingly popular across a range of disciplines. Leveraging natural language processing (NLP) techniques enables the analysis of textual data in novel ways and facilitates the use of social media data in disaster management. Analyzing text from social media with NLP can enhance situational awareness, accelerate information dissemination, and monitor affected communities, which is crucial for government agencies and emergency decision makers. However, a comprehensive literature review of NLP and social media data in disaster management is lacking, which presents an obstacle to rapid development in this field. To address this gap, this project surveys 324 related articles published between 2011 and 2022 and investigates the current trends and state-of-the-art NLP applications for using social media data in managing natural disasters. The bibliometric analysis reveals that existing literature focused more on earthquakes, floods, and hurricanes, concentrating on during-event periods, with Twitter (now referred to as X) as the most cited information source. Moreover, our systematic analysis identifies common NLP methodologies and identifies five major applications in current research. Finally, this study provides important insights and potential directions for future research in social media, NLP, and disaster management.

Networks and identity drive the spatial diffusion of linguistic innovation in urban and rural areas

Npj Complexity, September 2024 

Aparna Ananthasubramaniam, David Jurgens, Daniel M. Romero 

Cultural innovation (e.g., music, beliefs, language) tends to be adopted regionally. The geographic area where innovation is adopted is often attributed to one of two factors: (i) speakers adopting new behaviors that signal their demographic identities (i.e., an identity effect), or (ii) these behaviors spreading through homophilous networks (i.e., a network effect). In this study, we show that network and identity play complementary roles in determining where new language is adopted; thus, modeling the diffusion of lexical innovation requires incorporating both network and identity. We develop an agent-based model of cultural adoption, and validate geographic properties in our simulations against a dataset of innovative words that we identify from a 10% sample of Twitter (e.g., fleeky, birbs, ubering). Using our model, we are able to directly test the roles of network and identity by comparing a model that combines network and identity against simulated network-only and identity-only counterfactuals. We show that both effects influence different mechanisms of diffusion. Specifically, network principally drives spread among urban counties via weak-tie diffusion, while identity plays a disproportionate role in transmission among rural counties via strong-tie diffusion. Diffusion between urban and rural areas, a key component in innovation spreading nationally, requires both network and identity. Our work suggests that models must integrate both factors in order to understand and reproduce the adoption of innovation.

Toward Integrated Takeover Performance Measurement: Validation of Fréchet Distance as a Takeover Performance Metric

Proceedings of the Human Factors and Ergonomics Society Annual Meeting, August 2024

Doo Won Han, Jundi Liu, X. Jessie Yang, Alicia Romo, William Horrey, Dawn Tilbury, Feng Zhou,  Lionel Robert, Lisa Molnar

This study introduces and validates a new metric for assessing takeover performance in conditionally automated driving, using Fréchet Distance. Fréchet Distance is a measurement that measures the similarity between two separate curves. Thirtytwo participants took part in a simulated driving experiment. Employing a 2×2 within-subjects design, the study compared traditional takeover performance metrics, including takeover time, time to collision, and resulting acceleration, with Fréchet Distance. Analysis results revealed similar trends between traditional metrics and Fréchet Distance. These findings suggest that Fréchet Distance can effectively measure takeover performance by integrating spatial and temporal aspects.

Pre-prints, Working Papers, Articles, Reports, Workshops and Talks

VRCopilot: Authoring 3D Layouts with Generative AI Models in VR

arXiv, August 2024

Lei Zhang, Jin Pan, Jacob Gettig, Steve Oney

Immersive authoring provides an intuitive medium for users to create 3D scenes via direct manipulation in Virtual Reality (VR). Recent advances in generative AI have enabled the automatic creation of realistic 3D layouts. However, it is unclear how capabilities of generative AI can be used in immersive authoring to support fluid interactions, user agency, and creativity. We introduce VRCopilot, a mixed-initiative system that integrates pre-trained generative AI models into immersive authoring to facilitate human-AI co-creation in VR. VRCopilot presents multimodal interactions to support rapid prototyping and iterations with AI, and intermediate representations such as wireframes to augment user controllability over the created content. Through a series of user studies, we evaluated the potential and challenges in manual, scaffolded, and automatic creation in immersive authoring. We found that scaffolded creation using wireframes enhanced the user agency compared to automatic creation. We also found that manual creation via multimodal specification offers the highest sense of creativity and agency.

AltCanvas: A Tile-Based Image Editor with Generative AI for Blind or Visually Impaired People

arXiv, August 2024

Seonghee Lee, Maho Kohga, Steve Landau, Sile O’Modhrain, Hari Subramonyam

People with visual impairments often struggle to create content that relies heavily on visual elements, particularly when conveying spatial and structural information. Existing accessible drawing tools, which construct images line by line, are suitable for simple tasks like math but not for more expressive artwork. On the other hand, emerging generative AI-based text-to-image tools can produce expressive illustrations from descriptions in natural language, but they lack precise control over image composition and properties. To address this gap, our work integrates generative AI with a constructive approach that provides users with enhanced control and editing capabilities. Our system, AltCanvas, features a tile-based interface enabling users to construct visual scenes incrementally, with each tile representing an object within the scene. Users can add, edit, move, and arrange objects while receiving speech and audio feedback. Once completed, the scene can be rendered as a color illustration or as a vector for tactile graphic generation. Involving 14 blind or low-vision users in design and evaluation, we found that participants effectively used the AltCanvas’s workflow to create illustrations.

Characterizing Online Toxicity During the 2022 Mpox Outbreak: A Computational Analysis of Topical and Network Dynamics

arXiv, August 2024

Lizhou Fan, Lingyao Li, Libby Hemphill

Background: Online toxicity, encompassing behaviors such as harassment, bullying, hate speech, and the dissemination of misinformation, has become a pressing social concern in the digital age. Its prevalence intensifies during periods of social crises and unrest, eroding the sense of safety and community. Such toxic environments can adversely impact the mental wellbeing of those exposed and further deepen societal divisions and polarization. The 2022 Mpox outbreak, initially termed "Monkeypox" but subsequently renamed to mitigate associated stigmas and societal concerns, serves as a poignant backdrop to this issue. 

Objective: In this research, we undertake a comprehensive analysis of the toxic online discourse surrounding the 2022 Mpox outbreak. Our objective is to dissect its origins, characterize its nature and content, trace its dissemination patterns, and assess its broader societal implications, with the goal of providing insights that can inform strategies to mitigate such toxicity in future crises. 

Methods: We collected more than 1.6 million unique tweets and analyzed them from five dimensions, including context, extent, content, speaker, and intent. Utilizing BERT-based topic modeling and social network community clustering, we delineated the toxic dynamics on Twitter. 

Results: By categorizing topics, we identified five high-level categories in the toxic online discourse on Twitter, including disease (46.6%), health policy and healthcare (19.3%), homophobia (23.9%), politics (6.0%), and racism (4.1%). Across these categories, users displayed negativity or controversial views on the Mpox outbreak, highlighting the escalating political tensions and the weaponization of stigma during this infodemic. Through the toxicity diffusion networks of mentions (17,437 vertices with 3,628 clusters), retweets (59,749 vertices with 3,015 clusters), and the top users with the highest in-degree centrality, we found that retweets of toxic content were widespread, while influential users rarely engaged with or countered this toxicity through retweets. 

Conclusions: Our study introduces a comprehensive workflow that combines topical and network analyses to decode emerging social issues during crises. By tracking topical dynamics, we can track the changing popularity of toxic content online, providing a better understanding of societal challenges. Network dynamics spotlight key social media influencers and their intents, indicating that addressing these central figures in toxic discourse can enhance crisis communication and inform policy-making.

Trust Dynamics in Human-Autonomy Interaction: Uncover Associations between Trust Dynamics and Personal Characteristics

arXiv, September 2024

Hyesun Chung, X. Jessie Yang

While personal characteristics influence people’s snapshot trust towards autonomous systems, their relationships with trust dynamics remain poorly understood. We conducted a human-subject experiment with 130 participants performing a simulated surveillance task aided by an automated threat detector. A comprehensive pre-experimental survey collected data on participants’ personal characteristics across 12 constructs and 28 dimensions. Based on data collected in the experiment, we clustered participants’ trust dynamics into three types and assessed differences among the three clusters in terms of personal characteristics, behaviors, performance, and post experiment ratings. Participants were clustered into three groups, namely Bayesian decision makers, disbelievers, and oscillators. Results showed that the clusters differ significantly in seven personal characteristics: masculinity, positive affect, extraversion, neuroticism, intellect, performance expectancy, and high expectations. The disbelievers tend to have high neuroticism, and low performance expectancy. The oscillators tend to have higher scores in masculinity, positive affect, extraversion and intellect. We also found significant differences in the behaviors and post experiment ratings among the three groups. The disbelievers are the least likely to blindly follow the recommendations made by the automated threat detector. Based on the significant personal characteristics, we developed a decision tree model to predict cluster types with an accuracy of 70%.

RELATED

Subscribe to our free Research Roundup newsletter today!