Lin Qiu
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
Lin Qiu has a total of 74 co-authors. Here are the top co-authors by number of shared publications:
Han Lin
Agency for Science Technology and Research, SINGAPORE
10 shared publications- Angela Ka Yee Leung
Department of Psychology The Chinese University of Hong Kong Hong Kong SAR China
9 shared publications - Chi-Yue Chiu
Faculty of Social Science, The Chinese University of Hong Kong, Hong Kong, China
6 shared publications - Sarah Hian May Chan
Lee Kuan Yew Centre for Innovative Cities, Singapore University of Technology and Design, Singapore, Singapore
6 shared publications - Jiahui Lu
School of New Media and Communication, Tianjin University, Tianjin, China
5 shared publications - William Tov
Singapore Management University
5 shared publications
Latest Publications
- Open Access
- 26 June 2026
- Published by Masaryk University Press in Cyberpsychology: Journal of Psychosocial Research on Cyberspace
- Vol. 20  (3)
Beautifully diverse? Self-representations and diversity preferences in avatar-mediated virtual environments
While online platforms increasingly prioritize diversity by offering extensive avatar customization options, it remains unclear how the avatar features that users are attracted to impact their objective diversity and whether perceived avatar diversity raises user engagement. We conducted two studies using a simulated avatar customization website to examine how user customization of avatars impacts each feature’s objective diversity. Further, drawing from the hyperpersonal model and aesthetic appreciation theory, we present users with groups of avatars that varied in levels of perceived attractiveness and diversity and assess their likelihoods of joining the platform. Study 1 analyzed 136 majority White participants (age range: 21 – 63 years; M = 37.4, SD = 10.8; 49.3% female, 47.8% male, 0.7% non-binary or third gender; 2.2% missing), and study 2 analyzed 128 non-White participants (age range: 21 – 65 years; M = 35.2, SD = 10.2; 47.7% female, 47.7% male; 4.7% missing). The results revealed two types of avatar features that were sample-dependent: (1) aesthetically-normative features, which align with widely accepted attractiveness norms, reduce objective diversity; and (2) aesthetically-expressive features, which allow creative self-expression, increase objective diversity. Regardless of the sample demographic, perceived avatar diversity amplified the positive effect of attractive avatars on users’ likelihood of joining the platform. This research demonstrates the utility of integrating the hyperpersonal model and aesthetic appreciation theory in theorizing the interaction between perceived avatar attractiveness and diversity on user engagement. It also uncovers a socio-technical feedback loop by studying objective and perceived diversity simultaneously. Actionable insights are provided to address the ethical implications of our findings.
- 26 June 2026
- Published by Masaryk University Press in Cyberpsychology: Journal of Psychosocial Research on Cyberspace
- Vol. 20  (3)
Open AccessBeautifully diverse? Self-representations and diversity preferences in avatar-mediated virtual environments
While online platforms increasingly prioritize diversity by offering extensive avatar customization options, it remains unclear how the avatar features that users are attracted to impact their objective diversity and whether perceived avatar diversity raises user engagement. We conducted two studies using a simulated avatar customization website to examine how user customization of avatars impacts each feature’s objective diversity. Further, drawing from the hyperpersonal model and aesthetic appreciation theory, we present users with groups of avatars that varied in levels of perceived attractiveness and diversity and assess their likelihoods of joining the platform. Study 1 analyzed 136 majority White participants (age range: 21 – 63 years; M = 37.4, SD = 10.8; 49.3% female, 47.8% male, 0.7% non-binary or third gender; 2.2% missing), and study 2 analyzed 128 non-White participants (age range: 21 – 65 years; M = 35.2, SD = 10.2; 47.7% female, 47.7% male; 4.7% missing). The results revealed two types of avatar features that were sample-dependent: (1) aesthetically-normative features, which align with widely accepted attractiveness norms, reduce objective diversity; and (2) aesthetically-expressive features, which allow creative self-expression, increase objective diversity. Regardless of the sample demographic, perceived avatar diversity amplified the positive effect of attractive avatars on users’ likelihood of joining the platform. This research demonstrates the utility of integrating the hyperpersonal model and aesthetic appreciation theory in theorizing the interaction between perceived avatar attractiveness and diversity on user engagement. It also uncovers a socio-technical feedback loop by studying objective and perceived diversity simultaneously. Actionable insights are provided to address the ethical implications of our findings.
- Open Access
Beyond Instrumental and Substitutive Paradigms: Introducing Machine Culture as an Emergent Phenomenon in Large Language Models
Recent scholarship typically characterizes Large Language Models (LLMs) through either an \textit{Instrumental Paradigm} (viewing models as reflections of their developers' culture) or a \textit{Substitutive Paradigm} (viewing models as bilingual proxies that switch cultural frames based on language). This study challenges these anthropomorphic frameworks by proposing \textbf{Machine Culture} as an emergent, distinct phenomenon. We employed a 2 (Model Origin: US vs. China) $\times$ 2 (Prompt Language: English vs. Chinese) factorial design across eight multimodal tasks, uniquely incorporating image generation and interpretation to extend analysis beyond textual boundaries. Results revealed inconsistencies with both dominant paradigms: Model origin did not predict cultural alignment, with US models frequently exhibiting ``holistic'' traits typically associated with East Asian data. Similarly, prompt language did not trigger stable cultural frame-switching; instead, we observed \textbf{Cultural Reversal}, where English prompts paradoxically elicited higher contextual attention than Chinese prompts. Crucially, we identified a novel phenomenon termed \textbf{Service Persona Camouflage}: Reinforcement Learning from Human Feedback (RLHF) collapsed cultural variance in affective tasks into a hyper-positive, zero-variance ``helpful assistant'' persona. We conclude that LLMs do not simulate human culture but exhibit an emergent Machine Culture -- a probabilistic phenomenon shaped by \textit{superposition} in high-dimensional space and \textit{mode collapse} from safety alignment.Open AccessBeyond Instrumental and Substitutive Paradigms: Introducing Machine Culture as an Emergent Phenomenon in Large Language Models
Recent scholarship typically characterizes Large Language Models (LLMs) through either an \textit{Instrumental Paradigm} (viewing models as reflections of their developers' culture) or a \textit{Substitutive Paradigm} (viewing models as bilingual proxies that switch cultural frames based on language). This study challenges these anthropomorphic frameworks by proposing \textbf{Machine Culture} as an emergent, distinct phenomenon. We employed a 2 (Model Origin: US vs. China) $\times$ 2 (Prompt Language: English vs. Chinese) factorial design across eight multimodal tasks, uniquely incorporating image generation and interpretation to extend analysis beyond textual boundaries. Results revealed inconsistencies with both dominant paradigms: Model origin did not predict cultural alignment, with US models frequently exhibiting ``holistic'' traits typically associated with East Asian data. Similarly, prompt language did not trigger stable cultural frame-switching; instead, we observed \textbf{Cultural Reversal}, where English prompts paradoxically elicited higher contextual attention than Chinese prompts. Crucially, we identified a novel phenomenon termed \textbf{Service Persona Camouflage}: Reinforcement Learning from Human Feedback (RLHF) collapsed cultural variance in affective tasks into a hyper-positive, zero-variance ``helpful assistant'' persona. We conclude that LLMs do not simulate human culture but exhibit an emergent Machine Culture -- a probabilistic phenomenon shaped by \textit{superposition} in high-dimensional space and \textit{mode collapse} from safety alignment. - Open Access
- 28 April 2025
- Published by Springer Nature in Journal of Computational Social Science
- Vol. 8  (2) , 1-19
CEOs’ online personality and their company’s business strategy: openness to experience predicts diversifying merger and acquisition decisions
As a personality trait, openness to experience plays a central role in driving behaviors associated with exploration and innovation. Despite decades of extensive research, the expression of openness to experience in business strategy remains understudied. The current study addresses this gap by revealing that the openness in experience perceived from American and Chinese CEOs’ social media predicted their firm’s initiation of diversifying M&As, even after controlling for an extensive set of potential confounds. By replicating our findings across two culturally distinct samples, we highlight the role of openness in corporate strategic decision-making, and the usefulness of social media as a source for personality assessment.
- 28 April 2025
- Published by Springer Nature in Journal of Computational Social Science
- Vol. 8  (2) , 1-19
Open AccessCEOs’ online personality and their company’s business strategy: openness to experience predicts diversifying merger and acquisition decisions
As a personality trait, openness to experience plays a central role in driving behaviors associated with exploration and innovation. Despite decades of extensive research, the expression of openness to experience in business strategy remains understudied. The current study addresses this gap by revealing that the openness in experience perceived from American and Chinese CEOs’ social media predicted their firm’s initiation of diversifying M&As, even after controlling for an extensive set of potential confounds. By replicating our findings across two culturally distinct samples, we highlight the role of openness in corporate strategic decision-making, and the usefulness of social media as a source for personality assessment.
- Open Access
- 24 April 2025
- Published by Springer Nature in Virtual Reality
- Vol. 29  (2) , 1-15
Planting trees in the metaverse changes self-identity and improves pro-environmental interests and attitudes towards the physical world
- 24 April 2025
- Published by Springer Nature in Virtual Reality
- Vol. 29  (2) , 1-15
Open AccessPlanting trees in the metaverse changes self-identity and improves pro-environmental interests and attitudes towards the physical world
- 8 August 2024
- Published by Springer Nature
- p. 169-180
An Investigation into the Stress-Buffering Effects of Nature Virtual Backgrounds in Video Calls
- 8 August 2024
- Published by Springer Nature
- p. 169-180
An Investigation into the Stress-Buffering Effects of Nature Virtual Backgrounds in Video Calls
- Open Access
Evaluating Psychological Safety of Large Language Models
In this work, we designed unbiased prompts to systematically evaluate the psychological safety of large language models (LLMs). First, we tested five different LLMs by using two personality tests: Short Dark Triad (SD-3) and Big Five Inventory (BFI). All models scored higher than the human average on SD-3, suggesting a relatively darker personality pattern. Despite being instruction fine-tuned with safety metrics to reduce toxicity, InstructGPT, GPT-3.5, and GPT-4 still showed dark personality patterns; these models scored higher than self-supervised GPT-3 on the Machiavellianism and narcissism traits on SD-3. Then, we evaluated the LLMs in the GPT series by using well-being tests to study the impact of fine-tuning with more training data. We observed a continuous increase in the well-being scores of GPT models. Following these observations, we showed that fine-tuning Llama-2-chat-7B with responses from BFI using direct preference optimization could effectively reduce the psychological toxicity of the model. Based on the findings, we recommended the application of systematic and comprehensive psychological metrics to further evaluate and improve the safety of LLMs.
Open AccessEvaluating Psychological Safety of Large Language Models
In this work, we designed unbiased prompts to systematically evaluate the psychological safety of large language models (LLMs). First, we tested five different LLMs by using two personality tests: Short Dark Triad (SD-3) and Big Five Inventory (BFI). All models scored higher than the human average on SD-3, suggesting a relatively darker personality pattern. Despite being instruction fine-tuned with safety metrics to reduce toxicity, InstructGPT, GPT-3.5, and GPT-4 still showed dark personality patterns; these models scored higher than self-supervised GPT-3 on the Machiavellianism and narcissism traits on SD-3. Then, we evaluated the LLMs in the GPT series by using well-being tests to study the impact of fine-tuning with more training data. We observed a continuous increase in the well-being scores of GPT models. Following these observations, we showed that fine-tuning Llama-2-chat-7B with responses from BFI using direct preference optimization could effectively reduce the psychological toxicity of the model. Based on the findings, we recommended the application of systematic and comprehensive psychological metrics to further evaluate and improve the safety of LLMs.
- Open Access
- 1 January 2024
- Published by Association for Computational Linguistics (ACL)
- p. 1826-1843
Evaluating Psychological Safety of Large Language Models
Xingxuan Li, Yutong Li, Lin Qiu, Shafiq Joty, Lidong Bing. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.
- 1 January 2024
- Published by Association for Computational Linguistics (ACL)
- p. 1826-1843
Open AccessEvaluating Psychological Safety of Large Language Models
Xingxuan Li, Yutong Li, Lin Qiu, Shafiq Joty, Lidong Bing. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.
- Top Cited Papers
- 26 November 2021
- Published by Springer Nature in Virtual Reality
- Vol. 27  (4) , 3285-3300
Nature in virtual reality improves mood and reduces stress: evidence from young adults and senior citizens
Large populations worldwide have been deprived from nature experiences due to mass quarantines and lockdowns during the COVID-19 pandemic, and face a looming mental health crisis. Virtual reality offers a safe and practical solution to increase nature exposure. This research examined the effects of virtual nature using a within-subject design with young adults (Study 1) and senior citizens (Study 2). Results from the young adult sample showed that walking in a virtual forest reduced negative affect due to enhanced nature connectedness, and reduced stress measured by heart rate. Consistently, the senior citizen sample reported improved positive affect due to enhanced nature connectedness after the virtual nature walk. Our findings unveil the underlying mechanism of how virtual nature may improve psychological well-being and demonstrated how virtual nature can be used as an intervention to promote mental health.
- 26 November 2021
- Published by Springer Nature in Virtual Reality
- Vol. 27  (4) , 3285-3300
Top Cited PapersNature in virtual reality improves mood and reduces stress: evidence from young adults and senior citizens
Large populations worldwide have been deprived from nature experiences due to mass quarantines and lockdowns during the COVID-19 pandemic, and face a looming mental health crisis. Virtual reality offers a safe and practical solution to increase nature exposure. This research examined the effects of virtual nature using a within-subject design with young adults (Study 1) and senior citizens (Study 2). Results from the young adult sample showed that walking in a virtual forest reduced negative affect due to enhanced nature connectedness, and reduced stress measured by heart rate. Consistently, the senior citizen sample reported improved positive affect due to enhanced nature connectedness after the virtual nature walk. Our findings unveil the underlying mechanism of how virtual nature may improve psychological well-being and demonstrated how virtual nature can be used as an intervention to promote mental health.
- 1 April 2021
- Published by American Psychological Association (APA) in Psychology of Popular Media
- Vol. 10  (2) , 256-266
Unemployment rate predicts anger in popular music lyrics: Evidence from top 10 songs in the United States and Germany from 1980 to 2017.
- 1 April 2021
- Published by American Psychological Association (APA) in Psychology of Popular Media
- Vol. 10  (2) , 256-266
Unemployment rate predicts anger in popular music lyrics: Evidence from top 10 songs in the United States and Germany from 1980 to 2017.
- 28 February 2020
- Published by Oxford University Press (OUP)
Agent-Based Modeling in Political Decision Making
Political systems involve citizens, voters, politicians, parties, legislatures, and governments. These political actors interact with each other and dynamically alter their strategies according to the results of their interactions. A major challenge in political science is to understand the dynamic interactions between political actors and extrapolate from the process of individual political decision making to collective outcomes. Agent-based modeling (ABM) offers a means to comprehend and theorize the nonlinear, recursive, and interactive political process. It views political systems as complex, self-organizing, self-reproducing, and adaptive systems consisting of large numbers of heterogeneous agents that follow a set of rules governing their interactions. It allows the specification of agent properties and rules governing agent interactions in a simulation to observe how micro-level processes generate macro-level phenomena. It forces researchers to make assumptions surrounding a theory explicit, facilitates the discovery of extensions and boundary conditions of the modeled theory through what-if computational experiments, and helps researchers understand dynamic processes in the real-world. ABM models have been built to address critical questions in political decision making, including why voter turnouts remain high, how party coalitions form, how voters’ knowledge and emotion affect election outcomes, and how political attitudes change through a campaign. These models illustrate the use of ABM in explicating assumptions and rules of theoretical frameworks, simulating repeated execution of these rules, and revealing emergent patterns and their boundary conditions. While ABM has limitations in external validity and robustness, it provides political scientists a bottom-up approach to study a complex system by clearly defining the behavior of various actors and generate theoretical insights on political phenomena.
- 28 February 2020
- Published by Oxford University Press (OUP)
Agent-Based Modeling in Political Decision Making
Political systems involve citizens, voters, politicians, parties, legislatures, and governments. These political actors interact with each other and dynamically alter their strategies according to the results of their interactions. A major challenge in political science is to understand the dynamic interactions between political actors and extrapolate from the process of individual political decision making to collective outcomes. Agent-based modeling (ABM) offers a means to comprehend and theorize the nonlinear, recursive, and interactive political process. It views political systems as complex, self-organizing, self-reproducing, and adaptive systems consisting of large numbers of heterogeneous agents that follow a set of rules governing their interactions. It allows the specification of agent properties and rules governing agent interactions in a simulation to observe how micro-level processes generate macro-level phenomena. It forces researchers to make assumptions surrounding a theory explicit, facilitates the discovery of extensions and boundary conditions of the modeled theory through what-if computational experiments, and helps researchers understand dynamic processes in the real-world. ABM models have been built to address critical questions in political decision making, including why voter turnouts remain high, how party coalitions form, how voters’ knowledge and emotion affect election outcomes, and how political attitudes change through a campaign. These models illustrate the use of ABM in explicating assumptions and rules of theoretical frameworks, simulating repeated execution of these rules, and revealing emergent patterns and their boundary conditions. While ABM has limitations in external validity and robustness, it provides political scientists a bottom-up approach to study a complex system by clearly defining the behavior of various actors and generate theoretical insights on political phenomena.