Decision, Operations & Information Technologies Theses and Dissertations

Permanent URI for this collectionhttp://hdl.handle.net/1903/2761

Browse

Now showing 1 - 20 of 93

DEVELOPING MULTIMODAL LEARNING METHODS FOR VIDEO UNDERSTANDING
(2024) Sun, Mingwei; Zhang, Kunpeng; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
In recent years, the field of deep learning, with a particular emphasis on multimodal representation learning, has experienced significant advancements. These advancements are largely attributable to groundbreaking progress in areas such as computer vision, voice recognition, natural language processing, and graph network learning. This progress has paved the way for a multitude of new applications. The domain of video, in particular, holds immense potential. Video is often considered the most potent form of digital content for communication and the dissemination of information. The ability to effectively and efficiently comprehend video content could prove instrumental in a variety of downstream applications. However, the task of understanding video content presents numerous challenges. These challenges stem from the inherently unstructured and complex nature of video, as well as its interactions with other forms of unstructured data, such as text and network data. These factors contribute to the difficulty of video analysis. The objective of this dissertation is to develop deep learning methodologies capable of understanding video across multiple dimensions. Furthermore, these methodologies aim to offer a degree of interpretability, which could yield valuable insights for researchers and content creators. These insights could have significant managerial implications.In the first study, I introduce an innovative network based on Long Short-Term Memory (LSTM), enhanced with a Transformer co-attention mechanism, designed for the prediction of apparent emotion in videos. Each video is segmented into clips of one-second duration, and pre-trained ResNet networks are employed to extract audio and visual features at the second level. I construct a co-attention Transformer to effectively capture the interactions between the audio and visual features that have been extracted. An LSTM network is then utilized to learn the spatiotemporal information inherent in the video. The proposed model, termed the Sec2Sec Co-attention Transformer, outperforms several state-of-the-art methods in predicting apparent emotion on a widely recognized dataset: LIRIS-ACCEDE. In addition, I conduct an extensive data analysis to explore the relationships between various dimensions of visual and audio components and their influence on video predictions. A notable feature of the proposed model is its interpretability, which enables us to study the contributions of different time points to the overall prediction. This interpretability provides valuable insights into the functioning of the model and its predictions. In the second study, I introduce a novel neural network, the Multimodal Co-attention Transformer, designed for the prediction of personality based on video data. The proposed methodology concurrently models audio, visual, and text representations, along with their intra-relationships, to achieve precise and efficient predictions. The effectiveness of the proposed approach is demonstrated through comprehensive experiments conducted on a real-world dataset, namely, First Impressions. The results indicate that the proposed model surpasses state-of-the-art methods in performance while preserving high computational efficiency. In addition to evaluating the performance of the proposed model, I also undertake a thorough interpretability analysis to examine the contribution across different levels. The insights gained from the findings offer a valuable understanding of personality predictions. Furthermore, I illustrate the practicality of video-based personality detection in predicting outcomes of MBA admissions, serving as a decision support system. This highlights the potential importance of the proposed approach for both researchers and practitioners in the field. In the third study, I present a novel generalized multimodal learning model, termed VAN, which excels in learning a unified representation of \textbf{v}isual, \textbf{a}coustic, and \textbf{n}etwork cues. Initially, I utilize state-of-the-art encoders to model each modality. To augment the efficiency of the training process, I adopt a pre-training strategy specifically designed to extract information from the music network. Subsequently, I propose a generalized Co-attention Transformer network. This network is engineered to amalgamate the three distinct types of information and to learn the intra-relationships that exist among the three modalities, a critical facet of multimodal learning. To assess the effectiveness of the proposed model, I collect a real-world dataset from TikTok, comprising over 88,000 videos. Extensive experiments demonstrate that the proposed model surpasses existing state-of-the-art models in predicting video popularity. Moreover, I have conducted a series of ablation studies to attain a deeper comprehension of the behavior of the proposed model. I also perform an interpretability analysis to study the contributions of each modality to the model performance, leveraging the unique property of the proposed co-attention structure. This research contributes to the field by proffering a more comprehensive approach to predicting video popularity on short-form video platforms.
TWO ESSAYS ON THE ROLE OF INFORMATION TRANSPARENCY IN MARKETPLACE OPERATIONS
(2024) Jiang, Jane Yi; Elmaghraby, Wedad J.; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
This dissertation encompasses two studies on the crucial role of information within marketplace operations. Collaborating with two platforms, we deliver empirical evidence and offer prescriptive insights into how information is conveyed to and perceived by customers, and the consequent impacts on sellers and the marketplace at large.The first study analyzes the introduction of the novel blockchain tracing technology into an online grocery marketplace. Our findings indicate that credible supply chain transparency encourages consumers to more readily buy traced products, especially those that are handling-sensitive or offered in less-trusted markets. Consequently, adopting third-party sellers experienced an average monthly revenue increase of up to 23.4\%. By utilizing structural estimation to understand how consumers assess product attributes and quality, we highlight that consumer responses (and welfare effects) vary in sophistication and size based on their prior experience with the product category. Additionally, we establish that consumers deem blockchain-based. The second study analyzes the unintended transparency issue associated with the pricing structure of bundle discounts and its consequences on product purchases and returns. Our findings reveal that customers tend to overlook complex pricing structures, leading to impulsive buying and increased returns. Enhancing customer attentiveness of pricing can decrease the Retailer's return rates by 20.9\%. Moreover, improving customer attentiveness to pricing benefits retailers by enabling them to create more versatile bundle offers, further optimizing their sales strategy.
WHITHER WONDER WOMEN? ESSAYS ON GENDER DIVERSITY IN IT-ENABLED PROFESSIONAL AND CREATIVE DOMAIN
(2023) Wang, Yifei; Ramaprasad, Jui; Gopal, Anandasivam; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
Working towards equality and inclusion around gender and race in society is critically important. Despite the increasing number of conversations around these issues, more work is needed to evaluate the causes of unequal participation of men and women in organizations, markets, and economies. In particular, the lack of equity in terms of representation and participation within the important information technology (IT) sector has often been viewed as an ongoing problem. My dissertation focuses on this specific sector and explores potential remedies to enhance the participation and representation of women in specific segments of IT-enabled work, albeit in three different empirical contexts. In my first essay, I investigate the unequal participation of women in IT labor markets and whether they are less willing to compete for complex and risky IT projects. Through multiple experiments on technically trained students, I find that women in the IT industry are more willing to participate in bidding for riskier projects, and their bids are higher than those of men. My second essay studies the issue of unequal representation by women within the digital music industry, where inequitable representation has been clearly noted. Women artists are often faced with less attention, respect, and market share. The study shows that TikTok dance challenges offer a low-cost, effective way to promote artists and increase visibility. The challenges are particularly beneficial for women music artists. My third essay examines the intersection of gender and race in digital music consumption after Floyd's death. I explore whether music can raise awareness of social justice issues and the role of Black artists as sensemaking agents. I find that hip-hop listeners increased after Floyd's death, particularly in less racially diverse cities. Black artists received more listenership across all genres, but consumption was skewed towards Black men artists, highlighting the underrepresentation of women in Black-dominated music genres. Collectively, the findings from these studies in my dissertation will provide valuable theoretical contributions, practical insights, and actionable solutions to bind the gender gap and make the digital markets more diverse and inclusive.
ECONOMETRIC ANALYSIS OF BIKE SHARING SYSTEMS
(2022) Cao, Huan; Tunca, Tunay TT; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
I study the efficiency of the dockless bike sharing system, and how to utilize the operational decisions to improve system efficiency and profit. In the first chapter, I empirically analyze riders' economic incentives in a dockless bike sharing system and explore how to improve the efficiency of this business model. Specifically, I aim to answer three main questions: (i) What is the impact of the number of bicycles in the system on efficiency? (ii) How can bike relocation be best used to improve utilization? (iii) How do the efficiency of dockless and dock-based systems compare? To address these questions, I first build a microeconomic model of user decision-making in a dockless bike sharing system. I then use this model together with transaction-level data from a major dockless bike-sharing firm to structurally estimate the customer utility and demand parameters. Using this estimation in counterfactual analysis, we find that the company can decrease the bicycle fleet size by 40% while maintaining 90% of transactions, leading to estimated savings of $6.5 Million. We further find that a spatial bicycle rebalancing system based on our customer utility model can improve daily transactions by approximately 19%. Finally, we demonstrate that without bicycle redistribution, a smartly designed dock-based system can significantly outperform a dockless system. Our model provides a utility-based model that allows companies to estimate not only transactions, but also the time and location of lost potential demand, which can be used to make targeted improvements to the geographic bike distribution. It also allows managers to fine-tune bicycle fleet sizes and spatial rebalancing parameters. Further, our structural demand modeling can be used to improve the efficiency of dock-based systems by helping with targeted dock location decisions. In the second chapter, using data from a major dockless bike sharing system in Beijing, I study the subscription behavior and its relationship with the service level and price. Specifically, I develop an econometric model to study the subscription behavior for both existing subscribers and new sign-ups, respectively, and build up a functional relationship between the service level and system demand level and reveal the dynamics and interplays among subscription, system demand, and service level, which helps to recover the evolution of the number of subscribers over time. Based on all the estimation results and functional relationships, I then construct an empirical framework and straighten out the relationship between company profit and bicycle fleet size/subscription price. The counterfactual results show that the marginal benefit of deploying a larger bicycle fleet is decreasing, and the company should be cautious in determining and adjusting the bicycle fleet size. In addition, the examination demonstrates that the current price is too low, and raising the price properly can achieve about a 25% profit increase. The results also show the value of understanding riders' sensitivity to price and how to use it to better their operational decisions and accomplish better financial results. The counterfactual framework I proposed can be utilized in various policy evaluations and provides important insights and recommendations to the bikeshare companies and regulators.
The Effect of Perceived Attitude Similarity on Performance Ratings
(1983) Feren, Dena Beatrice; Carroll, Stephen J.; Digital Repository at the University of Maryland; University of Maryland (College Park, Md)
This research consists of a laboratory study designed to test the notion that variance in performance ratings can be accounted for by the perception of the rater that the persons/he is evaluating is attitudinally similar or dissimilar to himself or herself. Student subjects were led to believe that a certain manager either agreed or disagreed with them on a number of attitudinal issues. Subjects then viewed a videotaped performance of the manager conducting a performance review with one of his problem subordinates. Subjects were asked to rate his performance using two different rating instruments -- a trait rating scale and a Behavior Observation Scale -- and to indicate personal liking for the manager. Extent of attitude similarity was manipulated on two levels with a control group . That is, some subjects were led to believe that the ratee was attitudinal l y similar to self, others that the ratee was dissimilar to self, and a third group received no information about the ratee's attitudes. The ratee's performance was manipulated on three levels. Some subjects viewed only a high performin1; manager, others viewed a moderate performer, and a third group viewed a lo w performing manager. Three different vignettes were prepared to represent the three levels of performance. Finally, a hard-performance-data condition was included to test the robustness of the attitude similarity effect. Some subjects received hard performance data, in the form of bar graphs, that was consistent with the level of performance portrayed in their videotaped vignette (i.e., those viewing the low performer received hard data indicative of low performance). It was hypothesized that perceived attitude similarity would have its greatest effect when performance was moderate, and when subjects did not receive hard performance data. The results did not support these predictions. The effect of perceived attitude similarity on performance ratings was not significant under any of the experimental conditions. Perceived similarity had a small, but significant effect on attraction; however, level of performance accounted for a far greater proportion of variance in attraction measures than perceived similarity. It was concluded that the rating task in this experiment failed to create the conditions under which perceived similarity would be most likely to exert an influence on ratings. Specifically, the rating task was not sufficiently ambiguous for student raters.
THE PRICE OF FRESH AIR: ESSAYS ON THE INTERACTIVE EFFECTS OF TECHNOLOGY AND AIR POLLUTION ON ECONOMIC ACTIVITY
(2022) Jeong, Jaehoon; Gopal, Anandasivam; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
With recent dramatic industrialization around the world, air quality has become a global issue. In my dissertation, I investigate the effects of air pollution on the omni-channel retail business and public transportation.In the first essay, I study how diminished air quality affects the substitutive relationship between offline and online sales associated with a cosmetics retailer located in South Korea. I specifically test how air pollution may affect the actual demand that occurs during these promotion days across the offline and online channels. Interestingly, polluted air boosts online sales and online promotion effectiveness. Unexpectedly, air pollution is unlikely to hurt offline sales, and even increase offline sales and offline promotion effectiveness. I also find a notion of the inverted-U shaped reaction to the seriousness of polluted air consistently in offline sales. The second essay examines the effect of mobile nudges on behavioral changes focusing on public transportation ridership. I study the effect of air quality categories with easy-to-interpret user interface and air quality notification using a regression discontinuity design. I show that mobile nudges effectively help users make better decisions to protect themselves. My additional analyses suggest that the effect of mobile nudges may vary by schedule flexibility and travel purposes. I also observe adaptation behavior to air pollution over the years. In my third essay, I study the interaction between air pollution, channels, and product categories in the online retail context. Combined with environmental changes, differences in product characteristics and channel fit can create varied patterns of demand shift. Considering air pollution-driven shopping motivation, I examine how air pollution affects relative product category sales across mobile and PC channels. My results show that air pollution can increase mobile sales volume more than PC sales volume in urban areas. Also, air pollution creates a larger effect on skin care products and lower priced items than on makeup products and higher priced items accordingly. Overall, my dissertation suggests theoretical and practical implications for the business and social impacts of air pollution, which should aid decision-makers in formulating business and sound policy.
An Operations Management Framework to Improve Geographic Equity in Liver Transplantation
(2022) Akshat, Shubham; Raghavan, S.; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
In the United States (U.S.), on average three people die every day awaiting a liver transplant for a total of 1,133 lives lost in 2021. While 13,439 patients were added to the waiting list in 2021, only 9,236 patients received liver transplantation. To make matters worse, there is significant geographic disparity across the U.S. in transplant candidate access to deceased donor organs. The U.S. Department of Health and Human Services (HHS) is keen to improve transplant policy to mitigate these disparities. The deceased donor liver allocation policy has been through three major implementations in the last nine years, but yet the issue persists. This dissertation seeks to apply operations management models to (i) understand transplant candidate behavior, and (ii) suggest improvements to transplant policy that mitigate geographic disparity. In the first essay, we focus on reducing disparities in the organ supply to candidate demand (s/d) ratios across transplant centers. We develop a nonlinear integer programming model that allocates organ supply to maximize the minimum s/d ratios across all transplant centers. We focus on circular donation regions that address legal issues raised with earlier organ distribution frameworks. This enables reformulating our model as a set-partitioning problem and our proposal can be viewed as a heterogeneous donor circle policy. Compared to the current Acuity Circles policy that has fixed radius circles around donation locations, the heterogeneous donor circle policy greatly improves both the worst s/d ratio, and the range of s/d ratios. With the fixed radius policy of 500 nautical miles (NM) the s/d ratio ranges from 0.37 to 0.84 at transplant centers, while with the heterogeneous circle policy capped at a maximum radius of 500NM the s/d ratio ranges from 0.55 to 0.60, closely matching the national s/d ratio of 0.5983. Broader sharing of organs is believed to mitigate geographic disparity. Recent policies are moving towards broader sharing in principle. In the second essay, we develop a patient's dynamic choice model to analyze her strategic response to a policy change. First, we study the impact of the Share 35 policy, a variant of broader sharing introduced in 2013, on the behavioral change of patients at the transplant centers (i.e., change in their organ acceptance probability), geographic equity, and efficiency (transplant quality, offer refusals, survival benefit from a transplant, and organ travel distance). We find that sicker patients became more selective in accepting organs (acceptance probability decreased) under the Share 35 policy. Second, we study the current Acuity Circles policy and conclude that it would result in lower efficiency (more offer refusals and a lower transplant benefit) than the previous Share 35 policy. Finally, we show that broader sharing in its current form may not be the best strategy to balance geographic equity and efficiency. The intuition is that by indiscriminately enlarging the pool of supply locations from where patients can receive offers, they tend to become more selective, resulting in more offer rejections and less efficiency. We illustrate that the heterogeneous donor circles policy that equalizes the s/d ratios across geographies is better than Acuity Circles in achieving geographic equity at the lowest trade-off on efficiency metrics. The previous two essays demonstrate the benefit of equalizing the s/d ratios across geographies. In December 2018 the Organ Procurement and Transplantation Network (OPTN) Board of Directors approved the continuous distribution framework as the desired policy goal for all the organ allocation systems. In this framework, the waiting list candidates will be prioritized based on several factors, each contributing some points towards the total score of a candidate. The factors in consideration are medical severity, expected post-transplant outcome, the efficient management of organ placement, and equity. However, the respective weights for each of these potential factors are not yet decided. In the third essay, we consider two factors, medical severity and the efficient management of organ placement (captured using the distance between the donor hospital and transplant center), and we design an allocation policy that maximizes the geographic equity. We develop a mathematical model to calculate the s/d ratio of deceased-donor organs at a transplant center in a continuous scoring framework of organ allocation policy. We then formulate a set-partitioning optimization problem and test our proposals using simulation. Our experiments suggest that reducing inherent differences in s/d ratios at the transplant centers result in saving lives and reduced geographic disparity.
PRICING AND EMPLOYMENT ISSUES IN THE PROVISION OF RIDE SERVICES
(2022) Cao, Ziwei; Ball, Michael; Kannan, P.K.; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
Recent years have witnessed the emergence and dramatic growth of platform businesses. This dissertation addresses two challenges of significance to ride service companies: 1) it investigates promotion effects on two-sided platforms; and 2) it models platform pricing and staffing strategies under a hybrid employment mode.In the first chapter, I broadly discuss the new challenges faced by the ride service platforms in recent years and provide perspectives on related research questions. In the second chapter, I study the network effects of different promotion methods in two-sided markets. Using data from a transportation-service platform, I specify a structural model that quantifies the respective promotional effects for price discounts and service upgrades. The results show that the primary effect of price discounts is to increase demand within the same service tier, whereas upgrades have stronger stickiness effects and spillover effects. Based on the estimates, I calculate the return on investment (ROI) and find that the ROI for upgrades is higher than that for discounts. Our counterfactual analyses show that as the platform matures, the importance of upgrades increases while the importance of price discounts decreases. These results provide important managerial implications for platforms on how to design optimal promotions to grow their business. In the third chapter, I model an on-demand platform that adopts a hybrid employment mode. This work is motivated by the recent public debate over the status of drivers for the major ride- hailing platforms as contract workers. My hybrid employment environment includes both contractors and full-time employees who receive a benefits package. In the hybrid model with driver control, drivers have the flexibility to decide how long to work and consequently whether to be an employee or a Contractor. Those who work over a certain number of hours will be classified as employees and receive a benefits package. The platform is a profit-maximizer and decides the optimal price based on the required benefit amount. As the benefit amount increases, the platform's profit decreases, which is consistent with strong gig company opposition to providing benefits. Moreover, I show that higher benefits make consumers and full-time drivers better off but decrease part-time drivers' welfare as well as overall social welfare. I consider a second model to better balance the platform's profitability and drivers' welfare. Under the hybrid model with platform control, the platform hires a limited number of full-time employees while guaranteeing that a minimum proportion of all work will be fulfilled by those employees. In this way, the profit loss due to the required benefits is capped, making this alternative policy a potentially viable solution from the platform’s perspective.
DATA-DRIVEN ESSAYS: ROLE OF PRICING AND RETURNS
(2022) Hemmati, Sahar; Elmaghraby, Wedad J; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
In this dissertation, we identify leavers in retailers’ operations such as their shipping policy design as well as markdown strategy which can significantly impact their product returns and the costs associated with it. We study how a high free shipping thresholds can induce shoppers to place orders with an intention to return a part of the order later. Using an empirical approach, which exploits natural experiments, we confirm that high threshold and/or high shipping fees induce a substantial order padding behavior, which leads to lower sales revenue, after adjusting for returns. However, we find that such behavior can be largely alleviated with frictions to returns. We propose that retailers looking to design their shipping policy should correctly account for return environment features. We also explore the link between a retailer’s markdown pricing strategies and its impact on customers’ purchase and return behavior. We specifically study the distinction between regular price markdowns and bundle price markdowns and the key contributors to such distinctions, and how they contribute to sales and customers’ return behavior. Capturing this notion and the heterogeneous impact of bundle and regular discounts on merchandise with different substitutability or complementarity as well as correlation between items at return stage, we offer recommendations on how a retailer should approach the design of their markdown strategies.
ON THE IMPLICATIONS OF NEW POLICIES, MARQUEE SELLERS, AND GREEN NUDGES IN ONLINE SECONDARY MARKETS FOR DURABLE IT PRODUCTS: EVIDENCE FROM EMPIRICAL STUDIES
(2021) Alhauli, Abdullah; Gopal, Anandasivam; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
The rapid pace of product development in the IT sector has led to a volume surge of product returns, giving rise to critical environmental threats that can potentially have significantly adverse ecological effects. One possible avenue to mitigate these negative effects pertains to the establishment of robust secondary markets for these products, so that their useful life can be enhanced. My dissertation seeks to study multiple aspects aimed at enhancing the efficiency of online secondary markets for durable IT products, using economic and behavioral theories. The first essay examines the extent to which firm policies in the primary market mitigate inefficiencies caused by adverse selection in the secondary market for IT products. I find that policies implemented by firms in the primary market with respect to their products can have beneficial effects in addressing adverse selection in the secondary markets. The second essay studies how adding a marquee seller to a B2B secondary market platform for IT products affects other sellers, in terms of the prices they obtain for comparable products. I show that the entry of a marquee seller has a positive effect on the prices obtained by other sellers on the platform. I further show that this positive effect on final prices is moderated by bidders multi-homing activity, and their level of involvement in the marquee seller’s site. Finally, through behavioral experiments performed on Amazon MTurk, my third essay examines the extent to which the use of behavioral interventions, in the form of green nudges, can enhance the propensity of used IT products being purchased in the secondary market, thereby increasing the lifetime of these products. I find that the efficacy of using green nudges to impact consumer behavior depends on the kind of motivation (i.e., internal versus external motivation) the nudge is delivering. I further find that the effectiveness of green nudges can vary based upon product price and perceived quality, and consumer demographics and latent personalities. Collectively, the findings from these studies in my dissertation provide valuable theoretical as well as practical insights about the effectiveness of different mechanisms for enhancing the efficiency of online secondary markets for durable IT products.
EMPIRICAL INVESTIGATION OF USERS’ SUCCESSFUL STRATEGIES IN ONLINE PLATFORMS - EVIDENCE FROM CROWD-SOURCING AND SOCIAL MEDIA PLATFORMS
(2021) Lysyakov, Mikhail; Viswanathan, Siva; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
With the proliferation and constant growth of online platforms, there has been an increasing interest among academicians and practitioners to understand various aspects of these platforms, including the effective design of platforms, their governance and user engagement. This dissertation seeks to add to this stream of research by leveraging large-scale unstructured data and corresponding data analytics and econometric techniques to examine users’ strategies in online social media and crowdsourcing platforms and gain insights into factors that lead to successful outcomes. The first essay examines the content strategies of closely competing firms on Twitter with a focus on how the similarity/dissimilarity of their content strategies impacts their online outcomes. I find that firms that are more adept at leveraging higher-level social media affordances, such as interactivity, collaboration, and online contests to differentiate their content strategies experience better outcomes as compared to their closest rivals that only leverage the basic technological affordances of social media. The second essay examines successful strategies of users (designers) in a crowdsourcing platform wherein clients post contests to solicit design solutions for a monetary reward. This study uses state-of-the-art deep learning and image analysis techniques to examine the strategies of experienced and less-experienced designers in open contests where later-entrants can potentially leverage information spillovers from earlier design submissions within a contest. I find that while later-entrants typically leverage information spillovers from earlier submissions in a contest, only experienced designers who are able to integrate information from multiple highly-rated early submissions are more likely to be successful. The third essay examines users’ strategies in response to the introduction of an Artificial Intelligence system for logo design in an online crowdsourcing design platform. In analyzing what differentiates successful contestants from the others, I find that the successful contestants significantly increase focus (i.e., the number of re-submissions per contest) and increase the emotional content as well as the complexity of their designs, in response to the introduction of the AI system. Collectively, the findings from these studies add to our understanding of successful strategies in online platforms and provide valuable insights to theory and practice.
Evolution and Current Practices of Ride Services
(2021) Noh, Daehoon; Tunca, Tunay I; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
In the past decade we have seen a rapid transformation in the ride service market with the advent of decentralized Ride Hailing (RH) services (e.g., Uber, Lyft), who gained significant market share at the expense of traditional Vertically Integrated (VI) taxi companies.However, it is not clear what the endgame of this competition between these two models will be. What is more, the current development of driverless car technology, which leads to another form of Vertically Integrated service model, is posing further questions on how the industry will shape in the future. In the first chapter, we analyze this question by game-theoretically modeling entry and competition between decentralized Ride Hailing and Vertically Integrated ride services. First, by comparing monopoly models, we find that due to the advantage of centralization, the VI model is more efficient in increasing firm profits, reducing delays and increasing social welfare compared to the RH model. However, in competition, the RH firm predominantly reverses this disadvantage, and gains the upper hand in the market with lower delays, higher market share, and higher profits due to its flexibility in setting its supply compared to the VI firm, which projects the success of the ride-hailing firms compared to the traditional taxi model. Furthermore, the entry of the RH firm into the market always improves social welfare, while the entry of an inefficient VI firm may reduce industry service levels and surprisingly decrease social despite introducing competition. Our results suggest that entry of a costly self-driving car technology may in fact hurt the industry as a whole and social welfare in a market that is served predominantly by a ride-hailing company, and this technology should be approached carefully by the industry and the regulators. In the second chapter, we examine the effect of offering ride distance information to the drivers in the ride-hailing two-sided market.This is of our interest because currently, the drivers cannot observe the riders' destination. However the leading companies in the ride-hailing business such as Uber and Lyft are experimenting the effect of offering this information to the drivers, but the effect is yet uncertain. It may introduce efficiency, or it may aggravate the cherry picking behavior which will be detrimental to the firm. We develop a game theoretical model of (i) a ride-hailing firm where the drivers do not observe the riders' ride distance (the unobservable case), and (ii) where they can observe (the observable case). We compare the two cases to determine when it would benefit the ride-hailing firm to offer this information to the drivers. We also compare consumer surplus and social welfare to see if the firm's decision may be in conflict with the social planner's. Our main finding is that when consumers are patient, the drivers' cherry picking behavior introduces efficiency to the operations of the observable case, making it optimal for the firm to offer observability to the drivers. However as consumers become impatient, this cherry picking behavior becomes burdensome for the observable case, and thus the firm's optimal decision is to not give observability. Furthermore, under certain parameter regions, the firm's decisions are in conflict with the social planner's objective, thus the government may need to devise a policy to maximize social welfare.
Data Analytics and Mathematical Models to Facilitate Disease Prevention in the U.S.
(2020) Apergi, Lida Anna; Baras, John; Golden, Bruce; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
The U.S. is leading in healthcare expenditures worldwide, but health outcomes in the U.S. are not reflective of the level of spending. Prevention plays a crucial role in improving the health of individuals in the U.S., since it helps people live longer and healthier lives. Preventive services include actions that prevent diseases from ever occurring, detect diseases at an early stage, and manage diseases that have already been diagnosed. In this dissertation, we use data analytics and mathematical modeling techniques to better understand factors that influence disease prevention and help provide efficient solutions. In the first part of this dissertation, we study two problems of disease prevention at the public health level. First, we investigate the impact of state-level vaccination exemption policy and of the highly publicized Disneyland measles outbreak on MMR vaccination rates of young children. At the same time, we highlight the impact that the choice of socioeconomic factors can have on measurement results. We estimate the impact of these policies using multiple linear regression. Furthermore, we study the sensitivity of the results by examining a number of different approaches for the selection of socioeconomic control variables. Second, we utilize big data to estimate the additive cost of chronic diseases and study their cost patterns. We model the cost based on a cost hierarchy; that is, the cost of each condition is modeled as a function of the number of other more expensive chronic conditions the individual has. Using large scale claims data, we identify members that suffer from one or more chronic conditions and estimate their healthcare expenditures. Through our analysis, we categorize the chronic conditions into different expenditure groups based on the characteristics of their cost profiles. In the second part of this dissertation, we study two problems of disease prevention at the healthcare provider level, focusing in the area of cardiology. First, we study the adoption of conversational agent technology by patients with heart failure. Conversational agents can help patients with heart failure to manage their condition and provide frequent feedback to their healthcare providers. We analyze data from two studies, with each study focusing on a different type of conversational agent. We compare the two types of conversational agent technologies in terms of patient engagement, and investigate which patient characteristics are important in determining the patient engagement. Second, we tackle the problem of outpatient scheduling in the cardiology department of a large medical center. The outpatients have to go through a number of diagnostic tests and treatments before they can complete the final procedure. We develop an integer programming model to schedule appointments that are convenient for the outpatients by minimizing the number of visits that the patients have to make to the hospital and the time they spend waiting in the hospital. Furthermore, we investigate whether scheduling outpatients in groups can lead to better schedules for the patients.
Marriages Made in Silico: Essays on Social Norms, Technology Adoption, and Institutions in Online Matrimonial Matching Platforms
(2020) Karmegam, Sabari Rajan; Gopal, Anandasivam; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
Online matrimonial platforms have emerged as a way to take the highly institutionalized process of arranged marriages online while preserving the offline social, cultural, and gender norms. While there is a rich body of empirical work on online dating, the corresponding literature on online matrimonial platforms is sparse. My dissertation seeks to fill this gap. In my first essay, I look at mobile adoption's role in online matrimonial platforms' engagement and matching outcomes. The analysis shows that unlike the dating market where the market's transaction costs are eased by the ubiquity and personal nature of the mobile device for all users, here subgroups associated with strong endogamous preferences benefit with mobile adoption. My work extends the mobile ecosystem study to the societal context where institutional norms take precedence and influence mobile adoption outcomes. In my second essay, I study how the search frictions, social norms, and disempowerment that results from the gender skew in online matching platforms can be mitigated by using appropriate market design. I use a quasi-experimental methodology by relying on two interventions designed by the platform to reduce women's cognitive load. The interventions improved the overall well-being of women on platforms. My work here aims to increase awareness on the role platforms needs to play to improve women's well-being while ensuring that online platforms do not unravel. In my third essay, I look at whether the sanctity of institutional norms and traditional markers of status - involvement of multiple stakeholders through parental involvement and social norms related to endogamy and gender roles are retained in online matrimonial platforms. I find that "platformization" leads to institutional unbundling, with outcomes guided by more liberal ethos. This essay extends the platform literature on institutional contexts and shows that transition to online settings may not be seamless. My dissertation thus contributes to the literature on Information Systems by highlighting the need to consider the societal, cultural, and gender norms to further our understanding of the market design and technology adoption in highly institutionalized contexts.
Mechanism Designs to Mitigate Disparities in Online Platforms: Evidence from Empirical Studies
(2020) Mayya, Raveesh K; Viswanathan, Siva; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
With the rising ubiquity of online platforms, there is an increasing focus on platforms’ role in enabling fair exchanges between buyers and sellers. Traditionally, platforms have inbuilt mechanisms such as screening or upfront data-gathering disclosure that encourage transactions between unfamiliar participants. Since such mechanisms can introduce power disparities between different sides, platforms have enacted policy changes to fix the imbalance. Extant literature hasn’t studied the unintended consequences of such policy changes. My dissertation seeks to fill this gap by examining platforms’ decisions to enact policy/mechanism changes that level the playing field by decentralizing choices for different sides. Using empirical studies, my dissertation seeks to causally identify the impact of such changes on outcomes for participants as well as for the platform. The first essay in my dissertation examines the impact Airbnb’s decision to make screening optional. There is increasing evidence that two-way screening mechanism has been used as a tool by users on the platform to discriminate against some users on the other side. In making screening optional, I find that African American hosts and female hosts are more likely to forgo screening and they benefit the most (in terms of occupancy, price and/or ratings) from forgoing screening, indicating that making screening optional can serve as a useful mechanism in helping alleviate reverse discrimination of hosts by guests. The second essay studies platforms’ attempts to provide smartphone users with better choice over which sensitive information can mobile apps access. In particular, I examine the timing of mobile apps' decisions to upgrade to Android 6.0, which restricts the ability of mobile apps from seeking blanket permissions to sensitive user information at download, instead requiring them to request à la carte permissions at run-time. I find that apps that over-seek (access information that are non-essential to their functionality) sensitive information from users strategically delay upgrading to Android 6.0. However, these apps suffer popularity and reputational costs in the Android marketplace. Collectively, the findings in my dissertation provides valuable theoretical as well as practical insights about the welfare implications of choice decentralization on all sides in online platforms, not just the intended side.
STUDYING THE IMPACT OF GOVERNMENT PROGRAMS ON HEALTHCARE EFFICIENCY USING ECONOMETRIC MODELS
(2020) Ren, Ai; Golden, Bruce; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
In this dissertation, we evaluate certain government interventions in several ways. We investigate the impact of the Global Budget Revenue (GBR) program on Length of Stay (LOS) for inpatients in emergency departments (EDs) in Maryland. The GBR program was announced on January 1, 2014 and, as a government mandate, all Maryland hospitals that were not covered by the Total Patient Revenue program were required to participate. Around the same time, many states, including Maryland, adopted Medicaid expansion. To estimate the effect of GBR on LOS in Maryland hospitals, we conduct a difference-in-differences (DID) analysis and consider the impact of Medicaid expansion by using hospitals from West Virginia, Delaware, and Rhode Island, which also adopted Medicaid expansion at the same time, as the control group. We expand the GBR study by adding more controls and using a longer study period. We find that all results support the conclusion that GBR implementation has a negative impact on the time that Maryland inpatients spend in the ED and, the bigger the hospital, the longer the LOS. We conduct a DID analysis and investigate the impact of the GBR program on the wait time for inpatients from admit decision to the time of departure from the ED in Maryland using four control groups based on different assumptions. Our estimates imply that GBR has a negative impact on the wait time of inpatients in the ED. Finally, we provide a comprehensive literature review of articles that used a DID model and were published since 1990 in the top 30 emergency medicine journals listed by the Scimago Institutions Rankings. We show that the top journals in emergency medicine have become more likely to publish DID-related articles. In the sixth chapter, we examine the changes in the number of fatality reports before and after the implementation of the Severe Injury Reporting Program (SIRP) using a DID model. Our study suggests the Occupational Safety and Health Administration (OSHA) should expand its approved state programs so as to increase implementation of the SIRP on the state level rather than the federal level. This could relieve pressure on OSHA’s limited resources while maintaining its commitment to national workplace safety.
Design and Operations on the Supply Side of Online Marketplaces
(2020) Zhang, Wenchang; Elmaghraby, Wedad J; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
Online platforms like eBay, Upwork, Airbnb, and Uber have transformed their markets, and many more are about to emerge. The rise of platforms has become one of the predominant economic and social developments of our time. Moreover, it has created many opportunities and challenges for both practitioners and researchers. My dissertation focuses on the design and operations on the supply side of online marketplaces. In particular, I study supply-side levers (e.g., listing policy and information provision policy) in different marketplace context (e.g., auction marketplace and service platform), with the consideration of strategic behavior of market participants and various friction involved in transactions (e.g., participation cost, information asymmetry, and supply adjustment friction). The first essay investigates how a one-sided liquidation auction marketplace maximizes its revenue by managing the supply-side market thickness under an exogenous supply inflow. The second essay examines the operational impacts of service platforms’ information disclosure regarding service providers’ qualities and revealing their mechanisms. The last essay studies whether two-sided marketplaces benefit or suffer from sellers’ quantity competition under unanticipated demand shocks. We further show that marketplaces can maneuver the competition in favorable directions by manipulating the supply adjustment friction. Overall, the findings from the three essays show that marketplaces’ operational levers on the supply side have significant effects on the strategies of all participants, which impacts the marketplaces’ operational performance. The dissertation offers both theoretical insights on the mechanisms of the studied supply-side levers and practical implications on how these levers should be designed and implemented.
USING ARTIFICIAL INTELLIGENCE TO IMPROVE HEALTHCARE QUALITY AND EFFICIENCY
(2020) Wang, Weiguang; Gao, Guodong Gordon; Agarwal, Ritu; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
In recent years, artificial intelligence (AI), especially machine learning (ML) and deep learning (DL), has represented one of the most exciting advances in science. The performance of ML-based AI in many areas, such as computer vision, voice recognition, and natural language processing has improved dramatically, offering unprecedented opportunities for application in a variety of different domains. In the critical domain of healthcare, great potential exists for a broader application of ML to improve quality and efficiency. At the same time, there are substantial challenges in the development and implementation of AI in healthcare. This dissertation aims to study the application of state-of-the-art AI technologies in healthcare, ranging from original method development to model interpretation and real-world implementation. First, a novel DL-based method is developed to efficiently analyze the rich and complex electronic health record data. This DL-based approach shows promise in facilitating the analysis of real-world data and can complement clinical knowledge by revealing deeper insights. Both knowledge discovery and performance of predictive models are demonstrably boosted by this method. Second, a recurrent neural network (named LSTM-DL) is developed and shown to outperform all existing methods in addressing an important real-world question, patient cost prediction. A series of novel analyses is used to derive a deeper understanding of deep learning’s advantages. The LSTM-DL model consistently outperforms other models with nearly the same level of advantages across different subgroups. Interestingly, the advantage of the LSTM-DL is significantly driven by the amount of fluctuation in the sequential data. By opening the “black box,” the parameters learned during the training period are examined, and is it demonstrated that LSTM-DL’s ability to react to high fluctuation is gained during the training rather than inherited from its special architecture. LSTM-DL can also learn to be less sensitive to fluctuations if the fluctuation is not playing an important role. Finally, the implementation of ML models in real practice is studied. Since at its current stage of development, ML-based AI will most likely assistant human workers rather than replace them, it is critical to understand how human workers collaborate with AI. An AI tool was developed in collaboration with a medical coding company, and successfully implemented in the real work environment. The impact of this tool on worker performance is examined. Findings show that use of AI can significantly boost the work productivity of human coders. The heterogeneity of AI’s effects is further investigated, and results show that the human circadian rhythm and coder seniority are both significant factors in conditioning productivity gains. One interesting finding regarding heterogeneity is that the AI has its best effects when a coder is at her/his peak of performance (as opposed to other times), which supports the theory of human-AI complementarity. However, this theory does not necessarily hold true across different coders. While it could be assumed that senior coders would benefit more from the AI, junior coders’ productivity is found to improve more. A further qualitative study uncovers the underlying mechanism driving this interesting effect: senior coders express strong resistance to AI, and their low trust in AI significantly hinders them from realizing the AI’s value.
DATA-DRIVEN OPTIMIZATION AND STATISTICAL MODELING TO IMPROVE DECISION MAKING IN LOGISTICS
(2019) Sinha Roy, Debdatta; Golden, Bruce; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
In this dissertation, we develop data-driven optimization and statistical modeling techniques to produce practically applicable and implementable solutions to real-world logistics problems. First, we address a significant and practical problem encountered by utility companies. These companies collect usage data from meters on a regular basis. Each meter has a signal transmitter that is automatically read by a receiver within a specified distance using radio-frequency identification (RFID) technology. The RFID signals are discontinuous, and each meter differs with respect to the specified distance. These factors could lead to missed reads. We use data analytics, optimization, and Bayesian statistics to address the uncertainty. Second, we focus on an important problem experienced by delivery and service companies. These companies send out vehicles to deliver customer products and provide services. For the capacitated vehicle routing problem, we show that reducing route-length variability while generating the routes is an important consideration to minimize the total operating and delivery costs for a company when met with random traffic. Third, we address a real-time decision-making problem experienced in practice. In one application, routing companies participating in competitive bidding might need to respond to a large number of requests regarding route costs in a very short amount of time. In another application, during post-disaster aerial surveillance planning or using drones to deliver emergency medical supplies, route-length estimation would quickly need to assess whether the duration to cover a region of interest would exceed the drone battery life. For the close enough traveling salesman problem, we estimate the route length using information about the instances. Fourth, we address a practical problem encountered by local governments. These organizations carry out road inspections to decide which street segments to repair by recording videos using a camera mounted on a vehicle. The vehicle taking the videos needs to proceed straight or take a left turn to cover an intersection fully. Right turns and U-turns do not capture an intersection fully. We introduce the intersection inspection rural postman problem, a new variant of the rural postman problem involving turns. We develop two integer programming formulations and three heuristics to generate least-cost vehicle routes.
Essays on Business Analytics
(2019) Gu, Liyi; Ryzhov, Ilya O; Business and Management: Decision & Information Technologies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
With the rapidly increasing availability of business-related big data in recent years as well as the advancements in statistical and machine learning techniques, business analytics (BA) is becoming an essential practice to explain past events, predict future trends and optimize decision making. Using BA, the two essays in this dissertation aim to address some important questions in two emerging topics: humanitarian fleet management and social behaviors in online gaming industry. In the first essay, we analyse how vehicle management is carried out in a humanitarian setting. In humanitarian fleet management, the performance of purchase, assignment, and sales decisions is determined by dynamic interactions between the fleet composition, the time-varying and uncertain demands on the fleet, and the depreciation of the vehicles as they are exploited. We propose to evaluate purchase, assignment, and sales policies in a realistic simulation environment that directly models heterogeneous vehicle attributes and tracks their evolution over time. Using data from a large international humanitarian organization (LIHO), the simulator can identify the rationale behind seemingly ad-hoc decisions by field managers at LIHO. For instance, by selling vehicles later than LIHO recommends, managers are actually reducing their costs; similarly, managers decline to switch vehicles between mission types because the benefits to the operational cost turn out to be marginal at best. In the second essay, we conduct an empirical study of the relationship between social interaction and user engagement, retention, and purchase behavior, based on a high-resolution player-level dataset from a major international video game company for one of its premier titles. We engineer a set of features that characterize social behavior within the game, and link these behaviors to several measures of user engagement using statistical and econometric models. Our results show that user engagement is highly correlated with certain social dynamics; meanwhile, social interaction does not always translate to better retention rates or more purchases. In some cases, high dependence on a small set of friends is positively correlated with churn, indicating a tradeoff between engagement in one title and adoption of others. Early adopters are generally more responsive to the social experience than late adopters.

Browse

Recent Submissions