News & Events
Faculty and students from across campus gathered to discuss their most recent projects and findings.
The Social Media and Business Analytics Collaborative sponsored another Spring Research Symposium on May 14, 2015, at the Carlson School of Management. Faculty and students from across campus gathered to share research projects related to social media, social computing, and data analytics. Industry guests also attended. The event concluded with a keynote presentation by Jeff Hancock of Cornell University. Hancock discussed the intense reaction to one of his research projects involving the Facebook news feed, and what he learned in the aftermath.
Projects and presenters:
Firms use employment promotions for two purposes: as an incentive and for matching. However, these goals may conflict; the best salesperson may not make the best sales manager. We test this using data from 244 companies that subscribe to hosted sales performance management software to examine who firms promote, comparing these to who makes a good manager. We find that firms promote strongly on raw sales performance in the prior job, ignoring other observable factors that better predict future managerial success.
In two experiments, we examine the effects of employer reputation in an online labor market in which employers may decline to pay workers while keeping the work product. In the first experiment, a blinded worker performs tasks posted by employers with good, bad, or no online reputations. Results confirm that reputation provides information on task completion time and nonpayment, and thereby effective wage rates. In the second experiment, we create multiple employer identities endowed with different exogenously introduced reputations. We find that employers with good reputations attract workers at nearly twice the rate as those with bad reputations with no discernible difference in quality. We interpret the results through the lens of an equilibrium search model in which the threat of a bad reputation deters employers from the abuse of authority even in the absence of contractual protections of workers.
Over the past few years, U.S. policymakers have introduced reforms aimed at increasing the quality and decreasing the cost of health care delivery. Many reforms are premised on the idea that improvements can be made by increasing integration among disparate providers to enable more coordinated patient care. Using data from millions of Medicare billing claims and patient discharge abstracts, this project maps and analyzes dynamic networks of care teams within and across U.S. hospitals. The findings suggest that network configurations may play an important role in shaping health care cost and quality.
The goal of this project is to understand how children and teenagers use online video sharing platforms like YouTube and Vine to connect with friends and a broader audience. Knowing more about their current practices can help us design interfaces that better support children’s creativity while keeping safety and privacy at the forefront. There are two current threads in this work: One is understanding current practices and designing supporting software. The other is developing general tools for understanding large video sets, particularly through crowdsourcing video analysis.
While the field of human-computer interaction (HCI) has thoroughly investigated audio and visual computer-mediated communication (e.g., video chat), the potential role of touch interaction has remained largely overlooked. We present three in-progress prototypes and a future study to explore this challenge.
Exploratory search, in which a user investigates complex concepts, is cumbersome with today’s search engines. Atlasify leverages thematic cartography to help people explore unfamiliar and complex information spaces for general exploratory search. Through innovations in natural language processing and spatial computing (available in our WikiBrain software library), Atlasify is able to support exploratory search in geography, as well as domains like chemistry, history, and politics.
People suffering from substance abuse disorders, such as addiction and alcoholism, frequently seek online support to maintain abstinence and manage their recovery. We present three ongoing projects on understanding how people in recovery use online support and how technology can best help them in this endeavor.
Social media / viral advertising messages shared among consumers typically have two different sources: the advertiser as the message creator and a consumer sender (or sharer) as the distributor. This project examined the influence of sender trust and advertiser trust on advertising effects. Results demonstrate sender trust and advertiser trust have differential influences on advertising outcomes. Significant interaction effects show that if ads are shared by trusted consumers, the influence of advertiser trust on ad outcomes becomes non-significant or reduced, suggesting that ad messages shared by trusted consumers can overcome the handicap unknown and less trusted advertisers might have.
Massive growth in online social networking has revitalized academic interest in the power of social contagion as a force for individual and collective action. Recent literature has causally established that peer-effects are ‘at-work’ in the general population of online social networks. Having established causal peer effects, it becomes natural to ask, how can we create, perhaps even maximize, social contagion in spreading peer influence? We conduct a randomized field experiment to examine how one mechanism – offline word-of-mouth – can be triggered using economic incentives. Our design involves manipulations of how a monetary referral reward is shared between the inviter and the invitee: selfish reward (inviter gets all), equal reward (50-50 split), and generous reward (invitee gets all). The unique context of mobile social gaming allows us to measure offline WOM as a driver for the adoption of digital goods. Our results show that in the aggregate general population, the generous pro-social referral rewards dominate purely selfish schemes in creating offline word-of-mouth. The results can help inform design of effective referral reward schemes for viral adoption in the digital world.
We present recent research on Pinterest. We show topic distribution, factors that drive attention, and the gender gap on Pinterest. Furthermore, we use Pinterest as a case to show how the popular "login with Facebook" integration may enlarge the gender gap.
The GroupLens Research Group has been a leader in Recommender Systems research for more than 20 years. This poster provides a view of current research efforts, including algorithm and interface improvements and new forms of social data fusion. It also provides an overview of research platforms, datasets, and other tools available for students, researchers, and industry use.
Using big and unstructured data on user reported adverse events related to medical devices, along with data from complementary databases, we applied predictive analytics to build a modeling framework for signal detection to predict medical device recalls, and evaluate the existence of judgment bias in making such predictions. The study demonstrates that prediction of failures of high tech innovations-in-use is possible with sufficient lead time by analyzing big and unstructured data on user feedback on adverse events in the marketplace; and that the precision of such predictions can be significantly improved by accounting for time-varying covariates related to design, supply chain, and manufacturing. The study yields novel insights into why firms fail to detect market signals of failures of high tech innovations-in-use. In particular, the study (i) shows that judgment bias prompts firms to under-react or over-react to market signals in the form of user feedback on adverse events, and (ii) identifies factors – such as event severity and noise-to-signal ratio in the data – that influence firms to under-react or over-react.
We analyzed 12,000 posts from Facebook pages of 41 companies across 6 industries, and found that users actually posted more negative posts than positive ones, and the negative posts led to greater engagement (i.e., more likes and comments by other Facebook users). We also found that likes and comments were two forms of engagement with different antecedents. Our research advances understanding of consumer-generated word-of-mouth on new social media platforms and has practical implications for business social media strategy.
Business pages on Facebook aim to engage consumers and build brand equity across the globe. To be global, brands must use localization strategies. We analyzed user posts on global brands’ Facebook pages across four cultures: U.S., Mexico, Australia, and Singapore. We share preliminary results from a qualitative analysis of posts from Volkswagen’s pages in the U.S. and Mexico. Compared to their Mexican counterparts, American fans were more likely to showcase ownership of Volkswagen cars, less likely to post inquiries directed at the company, and more likely to converse with other users about social events such as car shows. We also discuss how the findings relate to cultural dimensions such as individualism-collectivism and vertical-horizontal, and practical implications on social media marketing and global brand management.
Time allocation is a challenging control problem: humans must integrate importance and goal availability with our ability to complete them flexibly in response to changing demands and opportunities. Drawing on and extending time allocation results in foraging theory, we develop a rational theory for task switching as an optimal time allocation process. We combine tasks that provide instantaneous rewards and those that have goals on completion together in a unified framework. Incorporating intrinsic motivation ideas in artificial intelligence, we predict human task engagement in a task-switching game.
Substance abuse disorders, such as addiction and alcoholism, affect millions of people worldwide. Positive psychology intervention, such as gratitude lists and reflecting on personal experiences, have been shown to have a positive impact on recovery from these disorders. We are in the process of developing a mobile app to help individuals and groups integrate positive psychology interventions into their daily habits.
The geographic nature of platforms like Uber and TaskRabbit distinguish these platforms from purely online markets and raises new, fundamental questions. We carried out a controlled study on TaskRabbit and Uber in the Chicago metropolitan area. Quantitative modeling and qualitative data show that people in certain types of neighborhoods (e.g., poorer) are disadvantaged. Namely, people in these neighborhoods have a harder time hiring workers (e.g., finding a Uber car to pick them up, getting a TaskRabbit “tasker” to do a job) and may pay more for the same tasks in some situations.
Wikipedia’s 400+ million visitors generate over 20 billion page views every month, and many algorithms in domains like artificial intelligence build upon Wikipedia as well. However, are the interests of site contributors, and therefore content, in alignment with readers? And how even is the representation across socioeconomic statuses? Our work finds that a large part of Wikipedia’s audience sees significantly lower quality content than they would if contributors aligned their efforts with the readership. We also find significant differences in coverage across population density, racial, and economic lines.
In user-generated content systems (e.g., Twitter, OpenStreetMap), data quality and quantity is much lower in low socio-economic status (SES) areas. One type of user-generated content system, mobile crowdsourcing (geolocated crowdsourcing tasks done with mobile devices), may be able to counteract these biases. Our mobile crowdsourcing system, FolkSource, enables participants to easily choose spatially-tied tasks to complete. What does it take for mobile crowdsourcing systems to mitigate these biases? Our work seeks to understand, using FolkSource, what types of motivations and tools we can use to counteract the biases that exist in user-generated content systems.
Extant research on online review generation has primarily focused on valence (i.e., what rating to give). However, volume (i.e., whether to write a review) remains largely unexamined. Using a unique data set, we examine peer effect on content production, measured at the message level, and find that it does exist and is attributed to characteristics of the receiver, the object and the message, and the relationship between sender and receiver. We show that not all friends have the same effect on content production. A user is more likely affected by friends who share relevant content and whom the user admires. The findings hold implications for marketers and others wishing to facilitate content generation and spread Word of Mouth.
Online peer-to-peer (P2P) lending is a new form of financial market where individuals can collectively lend money to a borrower via an online platform. The decentralized lending model on P2P platforms enables herding among lenders which in turns holds important implications for platform performance. This paper aims to make two contributions: First, we propose and test a new method for capturing the degree of herding. We then use the new measurement to study the relationship between herding and loan outcomes. Initial results suggest that herding is associated with lower loan default rates.
This symposium was sponsored by the Social Media and Business Analytics Collaborative, a University-wide research initiative coordinated by the Carlson School of Management and the College of Science and Engineering.