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. 

---

“Longitudinal Associations between Online Usage of Library-Licensed Content and Undergraduate Student Performance”

M | Library Deep Blue Documents, February 2023 

Felichism Kabo, Annaliese Paulson, Dorren Bradley, Kenneth J. Varnum, Stephanie Teasley

To better understand the longitudinal association between online usage of library-licensed content and short- and long-term student performance, we linked EZproxy logs to institutional university data to study how library usage impacts semester and cumulative GPAs. Panel linear mixed effects regression models indicate online library usage is significantly associated with semester and cumulative GPAs. The library usage effect is larger for semester GPA and varies by on- and off-campus residency. The effect on semester GPA is larger for off-campus students, while for cumulative GPA the effect is larger for on-campus students. Longitudinally linked library-institutional data offers key insights on the library’s value.

---

“How Much Space Has Been Explored? Measuring the Chemical Space Covered by Databases and Machine-Generated Molecules”

International Conference on Learning Representations, February 2023

Yutong Xie, Ziqiao Xu, Jiaqi Ma, Qiaozhu Mei

Forming a molecular candidate set that contains a wide range of potentially effective compounds is crucial to the success of drug discovery. While most databases and machine-learning-based generation models aim to optimize particular chemical properties, there is limited literature on how to properly measure the coverage of the chemical space by those candidates included or generated. This problem is challenging due to the lack of formal criteria to select good measures of the chemical space. In this paper, we propose a novel evaluation framework for measures of the chemical space based on two analyses: an axiomatic analysis with three intuitive axioms that a good measure should obey, and an empirical analysis on the correlation between a measure and a proxy gold standard. Using this framework, we are able to identify #Circles, a new measure of chemical space coverage, which is superior to existing measures both analytically and empirically. We further evaluate how well the existing databases and generation models cover the chemical space in terms of #Circles. The results suggest that many generation models fail to explore a larger space over existing databases, which leads to new opportunities for improving generation models by encouraging exploration.

---

“Online Harassment: Assessing Harms and Remedies”

Social media and Society, March 2023

Sarita Schoenebeck, Cliff Lampe, Penny Triệu

Online harassment refers to a wide range of harmful behaviors, including hate speech, insults, doxxing, and non-consensual image sharing. Social media platforms have developed complex processes to try to detect and manage content that may violate community guidelines; however, less work has examined the types of harms associated with online harassment or preferred remedies to that harassment. We conducted three online surveys with US adult Internet users measuring perceived harms and preferred remedies associated with online harassment. Study 1 found greater perceived harm associated with non-consensual photo sharing, doxxing, and reputational damage compared to other types of harassment. Study 2 found greater perceived harm with repeated harassment compared to one-time harassment, but no difference between individual and group harassment. Study 3 found variance in remedy preferences by harassment type; for example, banning users is rated highly in general, but is rated lower for non-consensual photo sharing and doxxing compared to harassing family and friends and damaging reputation. Our findings highlight that remedies should be responsive to harassment type and potential for harm. Remedies are also not necessarily correlated with harassment severity—expanding remedies may allow for more contextually appropriate and effective responses to harassment.

---

“‘It’s Your Finsta at the End of the Day . . . Kind of’: Understanding Emerging Adults’ Self-Presentational Changes on Secondary Accounts”

Social media and Society, March 2023 

Michelle Tao and Nicole B. Ellison

This study explores emerging adults’ reflections about how their secondary Instagram account (“Finsta” or Fake Instagram) self-presentation evolved over time, from account creation to their present use. Drawing on interview data collected from female emerging adults (N = 17) who had at least one Finsta and one Rinsta (i.e., Real Instagram), we use Emerging Adulthood as a frame for understanding how their content-sharing practices changed, reflecting their own development from adolescence to adulthood. Our participants described how their Instagram content-sharing behaviors evolved, reflecting their desire to be seen as more mature, especially in regard to managing negative self-presentational content. Furthermore, some participants described the difficulties introduced by context collapse associated with presenting to two distinct networks, both high school and college followers. Finally, and reflecting contemporary developments, we document how the “Making Rinsta Casual Again” trend and the COVID-19 pandemic influence users’ self-presentational behaviors on image-based social media platforms like Instagram.

---

“Coyotes, Caravans, and Connectivity: Digital Practices among Honduran Irregular Migrants”

The International Conference on Information & Communication Technologies and Development, June 2022 

Sylvia Darling, Kentaro Toyama 

As with irregular migrants elsewhere, Hondurans seeking to cross the border into the United States use digital technologies, particularly mobile phones. Based on participant observation and semi-

structured interviews in Honduras with 26 irregular migrants, we explore how they relate to digital devices at the onset of their journey and throughout. Contrary to studies on migrants’ experiences at or near their desired destination, we find that participants strongly prefer to use mobile phones to the extent allowed by their modes of migration – specifically, travel with human smugglers called coyotes, with caravans, or on their own. Irregular migrants encounter dangerous threats along their path, which they fear could be heightened by mobile phones. Nevertheless, despite their

concerns, we find that their adaptive strategies are irrelevant or ineffective in preventing deportation and extortion. These findings lead to recommendations that consider a broader digital ecology beyond personal devices to support migrants’ needs and dispel misinformation about the dangers of device access.

---

“Op-Ed: Supporting Democracy through Content-Neutral Social Media Policies”

Journal of Science Policy & Governance, March 2023

Christopher L. Quarles

The internet and social media carry vast amounts of new information every second. To make these flows manageable, platforms engage in content moderation, using algorithms and humans to decide which content to recommend and which to remove. These decisions have profound effects on our elections, democratic debate, and human well-being. The U.S. government cannot directly regulate these decisions due to the scale of the content and the First Amendment. Rather than focusing exclusively on whether or what content gets moderated, policy-makers should focus on ensuring that incentives and processes create an information infrastructure that can support a robust democracy. These policies are most likely to be content-neutral. Three content-neutral mechanisms are promising targets for policy: process, transparency, and de-amplification.

---

“The benefit of reflection prompts for encouraging learning with hints in an online programming course”

The Internet and Higher Education, March 2023

Heeryung Choi, Jelena Jovanovic , Oleksandra Poquet, Christopher Brooks, Srećko Joksimović, Joseph Jay Williams 

While giving learners hints is a commonly used scaffolding practice to facilitate learning, previous work questioned the effectiveness of hints. In this study, we examined if prompting learners to reflect along with receiving hints could improve learning outcomes, including immediate and delayed performance, perceived learning, and enjoyment. A field experiment was conducted in a four-week long online master’s degree course on data science where we compared two conditions: a condition with hints and a condition providing reflection prompts along with hints. Results showed that using hints with reflection prompts increased learner performance in delayed knowledge transfer tasks while also increasing learners’ perception of learning. The combination of reflection prompts and hints did not reduce learners’ enjoyment of the tasks, suggesting that the use of hints with reflection prompts is not only an intervention which can improve learning outcomes but is one which will be naturally adopted by learners.

---

“A Prompt Log Analysis of Text-to-Image Generation Systems”

arXiv, March 2023

Yutong Xie, Zhaoying Pan, Jinge Ma, Jie Luo, Qiaozhu Mei

Recent developments in diffusion models have unleashed the astonishing capabilities of text-to-image generation systems to synthesize high-quality images that are faithful to a given reference text, known as a "prompt." These systems, once released to the public, have immediately received tons of attention from researchers, creators, and common users. Despite the plenty of efforts to improve the underneath generative models, there is limited work on understanding the information needs of the real users of these systems, e.g., by investigating the prompts the users input at scale. In this paper, we take the initiative to conduct a comprehensive analysis of large-scale prompt logs collected from multiple text-to-image generation systems. Our work is analogous to analyzing the query log of Web search engines, a line of work that has made critical contributions to the glory of the Web search industry and research. We analyze over two million user-input prompts submitted to three popular text-to-image systems at scale. Compared to Web search queries, text-to-image prompts are significantly longer, often organized into unique structures, and present different categories of information needs. Users tend to make more edits within creation sessions, showing remarkable exploratory patterns. Our findings provide concrete implications on how to improve text-to-image generation systems for creation purposes.