De Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
De Liu has a total of 73 co-authors. Here are the top co-authors by number of shared publications:
Andrew B. Whinston
University of Texas at Austin - Department of Information, Risk and Operations Management
6 shared publicationsRadhika Santhanam
5 shared publications- Zhihong Ke
Clemson University
4 shared publications Jianqing Chen
University of Calgary, Haskayne School of Business, Calgary, Alberta T2N 1N4, Canada
3 shared publications- Zihong Huang
Rawls College of Business, Texas Tech University, Lubbock, Texas 79409
3 shared publications - Andy Tao Li
Faculty of Business for Science and Technology, School of Management, University of Science and Technology of China, Hefei 230052, China
2 shared publications
Latest Publications
- Top Cited Papers
- 1 December 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
- Vol. 36 (4) , 2358-2374
Artificial Intelligence (AI) Assistant in Online Shopping: A Randomized Field Experiment on a Livestream Selling Platform
Livestream selling is an innovative form of online shopping that supports real-time interactions between streamers and consumers. However, a key challenge remains: Streamers have limited capacity to answer individual inquiries, whereas shoppers expect fast, personalized responses. This study investigates whether an AI-powered streaming assistant can address this tension by providing interactive, chat-based support to help consumers access and process information. Through a large-scale randomized field experiment on a leading livestream selling platform, we find that the AI assistant increases sales by 3.00% and reduces product return rates by 12.55%. Our analysis suggests that the AI assistant helps consumers feel more informed and confident in their purchases, thereby reducing uncertainty. At the same time, the AI assistant can occasionally disrupt the consumers’ livestream experience. Overall, the benefits of uncertainty reduction outweigh the negative influence of interruptions. For platform managers and policymakers, these findings evidence the potential of AI technology to enhance online commerce. The AI assistant is particularly effective for high-uncertainty products and for streamers with large audiences, offering implications for strategic deployment. Our research provides actionable insights for integrating AI into livestream selling and other digital commerce scenarios where real-time, AI-powered support can facilitate both consumer satisfaction and business growth.
- 1 December 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
- Vol. 36 (4) , 2358-2374
Top Cited PapersArtificial Intelligence (AI) Assistant in Online Shopping: A Randomized Field Experiment on a Livestream Selling Platform
Livestream selling is an innovative form of online shopping that supports real-time interactions between streamers and consumers. However, a key challenge remains: Streamers have limited capacity to answer individual inquiries, whereas shoppers expect fast, personalized responses. This study investigates whether an AI-powered streaming assistant can address this tension by providing interactive, chat-based support to help consumers access and process information. Through a large-scale randomized field experiment on a leading livestream selling platform, we find that the AI assistant increases sales by 3.00% and reduces product return rates by 12.55%. Our analysis suggests that the AI assistant helps consumers feel more informed and confident in their purchases, thereby reducing uncertainty. At the same time, the AI assistant can occasionally disrupt the consumers’ livestream experience. Overall, the benefits of uncertainty reduction outweigh the negative influence of interruptions. For platform managers and policymakers, these findings evidence the potential of AI technology to enhance online commerce. The AI assistant is particularly effective for high-uncertainty products and for streamers with large audiences, offering implications for strategic deployment. Our research provides actionable insights for integrating AI into livestream selling and other digital commerce scenarios where real-time, AI-powered support can facilitate both consumer satisfaction and business growth.
- 1 October 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Management Science
- Vol. 71 (10) , 8865-8883
Economics of Social Media Fake Accounts
Amid the rise of the influencer economy, fake social media accounts have become prevalent on many social media platforms. Yet the problem of fake accounts is still poorly understood, and so is the effectiveness of coping strategies. This research models the ecosystem of fake accounts in an influencer economy and obtains insights on fake account purchasing behaviors, the impact of antifake efforts, and the roles of various contextual factors. We show that as the antifake effort increases, the equilibrium may transition from a “pooling” equilibrium, where a low-quality influencer buys fake accounts to mimic a high-quality one, to a “costly separating” equilibrium, where a high-quality influencer may buy fake accounts to prevent mimicry from a low-quality influencer, and to a “naturally separating” equilibrium where low- and high-quality influencers are separated without buying fake accounts. We find that increasing antifake efforts and increasing social media literacy may sometimes result in more fake accounts. A purely profit-driven platform always prefers a pooling equilibrium with zero antifake effort. As a platform puts more weight on consumer welfare, it may exert a positive effort to induce a separating equilibrium, but the platform’s preferred antifake effort tends to be lower than that of consumers. We also find that the platform sometimes prefers a lower social media literacy and a lower fake account base price, whereas consumers prefer the opposite. In contrast, improving the antifake technology level can benefit both the platform and consumers. Our main insights are applicable to scenarios with more influencer types and repeated interactions. This paper was accepted by D. J. Wu, information systems. Supplemental Material: The online appendix is available at https://doi.org/10.1287/mnsc.2022.02616 .
- 1 October 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Management Science
- Vol. 71 (10) , 8865-8883
Economics of Social Media Fake Accounts
Amid the rise of the influencer economy, fake social media accounts have become prevalent on many social media platforms. Yet the problem of fake accounts is still poorly understood, and so is the effectiveness of coping strategies. This research models the ecosystem of fake accounts in an influencer economy and obtains insights on fake account purchasing behaviors, the impact of antifake efforts, and the roles of various contextual factors. We show that as the antifake effort increases, the equilibrium may transition from a “pooling” equilibrium, where a low-quality influencer buys fake accounts to mimic a high-quality one, to a “costly separating” equilibrium, where a high-quality influencer may buy fake accounts to prevent mimicry from a low-quality influencer, and to a “naturally separating” equilibrium where low- and high-quality influencers are separated without buying fake accounts. We find that increasing antifake efforts and increasing social media literacy may sometimes result in more fake accounts. A purely profit-driven platform always prefers a pooling equilibrium with zero antifake effort. As a platform puts more weight on consumer welfare, it may exert a positive effort to induce a separating equilibrium, but the platform’s preferred antifake effort tends to be lower than that of consumers. We also find that the platform sometimes prefers a lower social media literacy and a lower fake account base price, whereas consumers prefer the opposite. In contrast, improving the antifake technology level can benefit both the platform and consumers. Our main insights are applicable to scenarios with more influencer types and repeated interactions. This paper was accepted by D. J. Wu, information systems. Supplemental Material: The online appendix is available at https://doi.org/10.1287/mnsc.2022.02616 .
- 1 March 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
- Vol. 36 (1) , 162-183
Enhancing User Privacy Through Ephemeral Sharing Design: Experimental Evidence from Online Dating
In the dynamic world of online dating, a key challenge faced by platforms is the cold-start problem, where newly matched users are hesitant to engage due to privacy concerns. Our solution, ephemeral sharing, addresses this by balancing privacy with the need for personal information sharing. This feature allows personal photos to disappear and become untraceable soon after being viewed, reassuring users about their privacy. We conducted a large-scale randomized experiment with more than 70,000 users to evaluate the impact of ephemeral sharing. The results are compelling: users who could share ephemeral photos were more likely to send personal images alongside with their matching request, especially those with human faces, leading to more matches and higher engagement. Significantly, this effect was more pronounced among users who are more sensitive to their privacy. Furthermore, ephemeral sharing was found to reduce users’ concerns related to data collection, dissemination, and identity misuse, thereby increasing the willingness to share personal information. This approach not only enhances user privacy but also stimulates more active engagement on the platform. For dating platforms and similar platforms, adopting ephemeral sharing can revolutionize user experience. It provides a strategic advantage by boosting user personal information sharing and enhancing privacy, crucial for maintaining meaningful communication in online dating. This feature represents a significant step forward in designing user-centric, privacy-conscious platforms.
- 1 March 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
- Vol. 36 (1) , 162-183
Enhancing User Privacy Through Ephemeral Sharing Design: Experimental Evidence from Online Dating
In the dynamic world of online dating, a key challenge faced by platforms is the cold-start problem, where newly matched users are hesitant to engage due to privacy concerns. Our solution, ephemeral sharing, addresses this by balancing privacy with the need for personal information sharing. This feature allows personal photos to disappear and become untraceable soon after being viewed, reassuring users about their privacy. We conducted a large-scale randomized experiment with more than 70,000 users to evaluate the impact of ephemeral sharing. The results are compelling: users who could share ephemeral photos were more likely to send personal images alongside with their matching request, especially those with human faces, leading to more matches and higher engagement. Significantly, this effect was more pronounced among users who are more sensitive to their privacy. Furthermore, ephemeral sharing was found to reduce users’ concerns related to data collection, dissemination, and identity misuse, thereby increasing the willingness to share personal information. This approach not only enhances user privacy but also stimulates more active engagement on the platform. For dating platforms and similar platforms, adopting ephemeral sharing can revolutionize user experience. It provides a strategic advantage by boosting user personal information sharing and enhancing privacy, crucial for maintaining meaningful communication in online dating. This feature represents a significant step forward in designing user-centric, privacy-conscious platforms.
- Open Access
- 1 January 2025
- Published by Elsevier in SSRN Electronic Journal
Interleaved Design For E-Learning: Theory, Design, And Empirical Findings
The rapid development of e-learning has drawn increasing attention to the issue of how learners' learning activities can be better structured using technologies. This study focuses on how to improve e-learning performance by optimizing the structuring of learning sessions from the perspective of interleaving (i.e., mixing different topics in a learning session). Following the design science paradigm, this study chooses cognitive load theory as the kernel theory and proposes a new interleaving design-related-interleaving-that populates an interleaved session with related topics as a way of reducing cognitive load during an interleaved session. Drawing on the theoretical predictions, we design and instantiate a personalized learning system with the related-interleaving strategy by fusing educational strategies and machine learning techniques. The results from a two-month field experiment confirm that related-interleaving outperforms non-interleaving and unrelated-interleaving. Our findings also reveal that compared with unrelated-interleaving, related-interleaving benefits weak learners more and thus helps reduce learning performance disparities. This study demonstrates how personalized e-learning systems can be further improved from the perspective of interleaving.
- 1 January 2025
- Published by Elsevier in SSRN Electronic Journal
Open AccessInterleaved Design For E-Learning: Theory, Design, And Empirical Findings
The rapid development of e-learning has drawn increasing attention to the issue of how learners' learning activities can be better structured using technologies. This study focuses on how to improve e-learning performance by optimizing the structuring of learning sessions from the perspective of interleaving (i.e., mixing different topics in a learning session). Following the design science paradigm, this study chooses cognitive load theory as the kernel theory and proposes a new interleaving design-related-interleaving-that populates an interleaved session with related topics as a way of reducing cognitive load during an interleaved session. Drawing on the theoretical predictions, we design and instantiate a personalized learning system with the related-interleaving strategy by fusing educational strategies and machine learning techniques. The results from a two-month field experiment confirm that related-interleaving outperforms non-interleaving and unrelated-interleaving. Our findings also reveal that compared with unrelated-interleaving, related-interleaving benefits weak learners more and thus helps reduce learning performance disparities. This study demonstrates how personalized e-learning systems can be further improved from the perspective of interleaving.
- Open Access
- 1 January 2025
- Published by Elsevier in SSRN Electronic Journal
Is ChatGPT a Boon or a Bane for Learning? Experimental Evidence Across Task Formats and Chatbot Designs
As concerns mount over the widespread use of large language model-based chatbots for schoolwork, our understanding remains limited regarding their effects on learning and how to customize LLM chatbots and learning tasks to promote effective learning in the era of generative AI. To address these issues, we conducted randomized field experiments among college students, randomly assigning each participant to one of three tools during a practice session: a Bing-based search engine, a standard ChatGPT chatbot, or a custom ChatGPT chatbot designed using the guided discovery principle through zero-shot prompting. Using subsequent paper-and-pencil exams to measure learning outcomes, we found that students with access to standard ChatGPT scored 4.18% lower than those with access to the search engine, despite ChatGPT being 3.8 times more likely to be used. Interestingly, the standard ChatGPT demonstrated negative effects primarily for selected-response problems, whereas positive effects were observed for constructed-response problems. In contrast, students provided with the guided-discovery ChatGPT scored 9.23% higher on exams and outperformed the standard ChatGPT on both selected-response and constructed-response problems, an outcome attributable to extended practice time, more conversation rounds, and greater critical engagement. Finally, post-survey responses revealed that students overestimated the learning benefits of the standard ChatGPT and relied on it more heavily, despite recognizing the advantages of guided-discovery ChatGPT, highlighting the risk of overreliance on standard LLM chatbots.
- 1 January 2025
- Published by Elsevier in SSRN Electronic Journal
Open AccessIs ChatGPT a Boon or a Bane for Learning? Experimental Evidence Across Task Formats and Chatbot Designs
As concerns mount over the widespread use of large language model-based chatbots for schoolwork, our understanding remains limited regarding their effects on learning and how to customize LLM chatbots and learning tasks to promote effective learning in the era of generative AI. To address these issues, we conducted randomized field experiments among college students, randomly assigning each participant to one of three tools during a practice session: a Bing-based search engine, a standard ChatGPT chatbot, or a custom ChatGPT chatbot designed using the guided discovery principle through zero-shot prompting. Using subsequent paper-and-pencil exams to measure learning outcomes, we found that students with access to standard ChatGPT scored 4.18% lower than those with access to the search engine, despite ChatGPT being 3.8 times more likely to be used. Interestingly, the standard ChatGPT demonstrated negative effects primarily for selected-response problems, whereas positive effects were observed for constructed-response problems. In contrast, students provided with the guided-discovery ChatGPT scored 9.23% higher on exams and outperformed the standard ChatGPT on both selected-response and constructed-response problems, an outcome attributable to extended practice time, more conversation rounds, and greater critical engagement. Finally, post-survey responses revealed that students overestimated the learning benefits of the standard ChatGPT and relied on it more heavily, despite recognizing the advantages of guided-discovery ChatGPT, highlighting the risk of overreliance on standard LLM chatbots.
- Open Access
- 1 January 2025
- Published by HICSS Conference Office
Social Media Impersonation and Verification Badge
Due to the increasing prevalence of impersonation on social media, the platforms have actively adopted verification badges to combat such an issue. To explore whether providing verification badges is an appropriate approach, we build a game-theoretic model to examine the impact of providing such badges on key stakeholders' payoffs. Our study reveals that offering verification badges can lead to unintended consequences. For example, we find that the platform encourages both creators and impersonators to purchase badges when the verification cost is low while discouraging badge application when the preparation cost is high. Surprisingly, providing verification badges can disadvantage the genuine creator by enhancing impersonator credibility, making it harder for consumers to distinguish between real and fake profiles, ultimately benefiting impersonators more. Additionally, consumers may be worse off when the influencers purchase the badge, as verified yet fraudulent accounts can more effectively deceive users, reducing the overall utility of consuming content.
- 1 January 2025
- Published by HICSS Conference Office
Open AccessSocial Media Impersonation and Verification Badge
Due to the increasing prevalence of impersonation on social media, the platforms have actively adopted verification badges to combat such an issue. To explore whether providing verification badges is an appropriate approach, we build a game-theoretic model to examine the impact of providing such badges on key stakeholders' payoffs. Our study reveals that offering verification badges can lead to unintended consequences. For example, we find that the platform encourages both creators and impersonators to purchase badges when the verification cost is low while discouraging badge application when the preparation cost is high. Surprisingly, providing verification badges can disadvantage the genuine creator by enhancing impersonator credibility, making it harder for consumers to distinguish between real and fake profiles, ultimately benefiting impersonators more. Additionally, consumers may be worse off when the influencers purchase the badge, as verified yet fraudulent accounts can more effectively deceive users, reducing the overall utility of consuming content.
- 1 June 2024
- Published by Springer Nature
- p. 259-264
Design and Evaluation of a Gamified Generative AI Chatbot for Canvas LMS Courses
- 1 June 2024
- Published by Springer Nature
- p. 259-264
Design and Evaluation of a Gamified Generative AI Chatbot for Canvas LMS Courses
- 8 May 2024
- Published by SAGE Publications in Production and Operations Management
- Vol. 34 (12) , 3998-4014
Magnify Cascades of Electronic Word-of-Mouth (eWOM) on Social Networks: The Roles of User, Product, and Relationship Characteristics
Firms are increasingly relying on electronic word-of-mouth (eWOM), in the form of online reviews and social media posts, to sell their products or services. A pivotal component of managing eWOM is to understand and, potentially influence, how one consumer's eWOM can lead to another's—a process called behavioral cascading . While prior eWOM research has established the importance of behavioral cascading between connected users, there is little understanding of what factors could impact the likelihood of such cascades. To address this gap, we draw on the theory of competitive altruism to identify several moderators of behavioral cascading in eWOM. Our empirical tests using an online review dataset from Yelp show that eWOM cascading between a followee and a follower is less likely when the followee is a high-status member, a female, or has a strong connection with the follower; and more likely when the product in consideration is inexpensive. These findings provide valuable insights about the behavioral cascading process in eWOM and hold implications for social media platforms and sellers to facilitate the cascades of eWOM between connected consumers.
- 8 May 2024
- Published by SAGE Publications in Production and Operations Management
- Vol. 34 (12) , 3998-4014
Magnify Cascades of Electronic Word-of-Mouth (eWOM) on Social Networks: The Roles of User, Product, and Relationship Characteristics
Firms are increasingly relying on electronic word-of-mouth (eWOM), in the form of online reviews and social media posts, to sell their products or services. A pivotal component of managing eWOM is to understand and, potentially influence, how one consumer's eWOM can lead to another's—a process called behavioral cascading . While prior eWOM research has established the importance of behavioral cascading between connected users, there is little understanding of what factors could impact the likelihood of such cascades. To address this gap, we draw on the theory of competitive altruism to identify several moderators of behavioral cascading in eWOM. Our empirical tests using an online review dataset from Yelp show that eWOM cascading between a followee and a follower is less likely when the followee is a high-status member, a female, or has a strong connection with the follower; and more likely when the product in consideration is inexpensive. These findings provide valuable insights about the behavioral cascading process in eWOM and hold implications for social media platforms and sellers to facilitate the cascades of eWOM between connected consumers.
- 1 March 2024
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
- Vol. 35 (1) , 272-293
Better to Give Than to Receive: Impact of Adding a Donation Scheme to Reward-Based Crowdfunding Campaigns
The choice funding scheme is an important design dimension in crowdfunding for which platforms should not assume that they are restricted to the use of one funding scheme. In particular, we find that the addition of a donation scheme can increase the success rate of reward campaigns substantially, especially those with prosocial causes. In addition, the donation received is found to have a crowd-in effect on subsequent contributions. To get the most out of donation schemes, crowdfunding sites should provide features that help backers locate reward campaigns for prosocial causes, for example, by providing tags or search filters related to prosocial causes. Second, campaign owners should take steps to highlight the prosocial aspects of their crowdfunding campaign, for example, by revising campaign descriptions to include prosocial keywords/tags. Third, campaign owners should play a more active role in encouraging early donations to enjoy the crowd-in effect that could help their campaigns reach the targeted funding amount. Specifically, the owner’s direct social network is a potentially good source for soliciting these early donations. Relatedly, the campaign owners could benefit by spreading the word about the early donations they receive so as to generate a greater crowd-in effect at later phases of the crowdfunding process.
- 1 March 2024
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
- Vol. 35 (1) , 272-293
Better to Give Than to Receive: Impact of Adding a Donation Scheme to Reward-Based Crowdfunding Campaigns
The choice funding scheme is an important design dimension in crowdfunding for which platforms should not assume that they are restricted to the use of one funding scheme. In particular, we find that the addition of a donation scheme can increase the success rate of reward campaigns substantially, especially those with prosocial causes. In addition, the donation received is found to have a crowd-in effect on subsequent contributions. To get the most out of donation schemes, crowdfunding sites should provide features that help backers locate reward campaigns for prosocial causes, for example, by providing tags or search filters related to prosocial causes. Second, campaign owners should take steps to highlight the prosocial aspects of their crowdfunding campaign, for example, by revising campaign descriptions to include prosocial keywords/tags. Third, campaign owners should play a more active role in encouraging early donations to enjoy the crowd-in effect that could help their campaigns reach the targeted funding amount. Specifically, the owner’s direct social network is a potentially good source for soliciting these early donations. Relatedly, the campaign owners could benefit by spreading the word about the early donations they receive so as to generate a greater crowd-in effect at later phases of the crowdfunding process.
- Open Access
- 1 January 2023
- Published by Association for Information Systems in Journal of the Association for Information Systems
- Vol. 24 (1) , 136-160
The Effects of Digitally Delivered Nudges in a Corporate Wellness Program
We investigate how two digitally delivered nudges, namely light social support (nonverbal cues such as kudos or likes) and motivational messaging, affect employees’ self-reported physical activity in an online, corporate wellness program. Within this unique field setting, using data from several years, we found evidence that both types of nudges provide benefits beyond the effect of cash incentives. However, the effects vary by individual, depending on whether the employee is actively engaging in physical activity, and by time, depending on how long the employee has been in the wellness program. We found light social support to be less effective over time, while motivational messages were found to be more effective with the duration in the program and generally more effective for physically inactive users. Our findings have implications for the design of wellness systems, suggesting different approaches depending on an employee’s current activity level and tenure in the program
- 1 January 2023
- Published by Association for Information Systems in Journal of the Association for Information Systems
- Vol. 24 (1) , 136-160
Open AccessThe Effects of Digitally Delivered Nudges in a Corporate Wellness Program
We investigate how two digitally delivered nudges, namely light social support (nonverbal cues such as kudos or likes) and motivational messaging, affect employees’ self-reported physical activity in an online, corporate wellness program. Within this unique field setting, using data from several years, we found evidence that both types of nudges provide benefits beyond the effect of cash incentives. However, the effects vary by individual, depending on whether the employee is actively engaging in physical activity, and by time, depending on how long the employee has been in the wellness program. We found light social support to be less effective over time, while motivational messages were found to be more effective with the duration in the program and generally more effective for physically inactive users. Our findings have implications for the design of wellness systems, suggesting different approaches depending on an employee’s current activity level and tenure in the program