Yi Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yi Wang has a total of 38 co-authors. Here are the top co-authors by number of shared publications:
- Thuong Hoang
School of Information Technology, Deakin University, Melbourne, Australia
6 shared publications - Chetan Arora
Monash University, Melbourne, Australia
5 shared publications - Xiao Liu
Deakin University, Melbourne, VIC, Australia
4 shared publications - John Grundy
Monash University, Melbourne, Australia
3 shared publications - Dan Zhang
Faculty of Science, Engineering and Built Environment, School of Information Technology, Deakin University, Burwood, VIC 3125, Australia
2 shared publications - Henry Been-Lirn Duh
School of Design, The Hong Kong Polytechnic University, Hong Kong, China
2 shared publications
Latest Publications
- Open Access
- 9 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Forest Classroom: Conversational AR Avatar for Help-Seeking
- 9 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Open AccessForest Classroom: Conversational AR Avatar for Help-Seeking
- Open Access
- 6 October 2026
- Published by Elsevier in SSRN Electronic Journal
PriVis: Proactive and Explainable Visualization of Visual Privacy Inferences on AI-Powered AR Glasses
With the rapid advancement of AI-powered augmented reality (AR) glasses, application providers can use Vision-Language Models (VLMs) to derive sensitive user profiles through latent privacy inferences. However, techniques to mitigate such risks remain underexplored. We present PriVis, an explainable visualization framework for AI-powered AR glasses designed to enhance user awareness and support informed control over privacy inferences derived from environmental objects. PriVis identifies source objects, infers private attributes, and presents explanations that quantify inference likelihood and the contribution of individual objects. A user study (N=24) evaluated PriVis against alternative conditions lacking explainable visualization or user control. Results show that PriVis outperformed baselines in terms of privacy awareness and control.
- 6 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPriVis: Proactive and Explainable Visualization of Visual Privacy Inferences on AI-Powered AR Glasses
With the rapid advancement of AI-powered augmented reality (AR) glasses, application providers can use Vision-Language Models (VLMs) to derive sensitive user profiles through latent privacy inferences. However, techniques to mitigate such risks remain underexplored. We present PriVis, an explainable visualization framework for AI-powered AR glasses designed to enhance user awareness and support informed control over privacy inferences derived from environmental objects. PriVis identifies source objects, infers private attributes, and presents explanations that quantify inference likelihood and the contribution of individual objects. A user study (N=24) evaluated PriVis against alternative conditions lacking explainable visualization or user control. Results show that PriVis outperformed baselines in terms of privacy awareness and control.
- Open Access
- 25 August 2026
- Published by Springer Nature in Virtual Reality
- Vol. 31 (1)
A systematic review of emotional nonverbal cues in embodied virtual agents for virtual environments
Embodied virtual agents (EVAs) are increasingly deployed in immersive virtual environments, where their nonverbal cues serve as a primary channel for affective communication. However, systematic evidence on how these nonverbal cues shape users’ emotional perception in virtual environments remains limited and dispersed across the literature. This paper presents a systematic literature review from a human-perception perspective to examine EVAs’ nonverbal cues in virtual environments and their effects on users’ affective perception. Searches were conducted in Scopus, the ACM Digital Library, IEEE Xplore, Web of Science, and SpringerLink, yielding 1066 records. Following the PRISMA guidelines, 52 studies met the inclusion criteria. The results indicate that facial expressions, posture and gaze are the most commonly used cues that influence users’ affective outcomes, with postural cues often demonstrating greater effectiveness in virtual environments. EVAs’ nonverbal expressions shape users’ affective perception across multiple dimensions, including their emotional responses, perceived social affect, and sense of social presence. More importantly, the immersive properties of virtual environments amplify the emotional contagion effects of these cues, while also increasing users’ sensitivity to subtle inconsistencies. Furthermore, the effectiveness of EVAs’ nonverbal cues in shaping users’ affective perceptions is context-dependent, with differing priorities across application domains. Advances in nonverbal generation for EVAs, from rule-based systems to generative AI, have progressively improved fidelity and affective perception, although they may also give rise to the uncanny valley effect. This review summarizes that EVAs’ nonverbal emotional cues substantially shape users’ affective perception in virtual environments, underscoring the importance of coherent multimodal cue design and context-appropriate generation methods for creating emotionally effective virtual agents.
- 25 August 2026
- Published by Springer Nature in Virtual Reality
- Vol. 31 (1)
Open AccessA systematic review of emotional nonverbal cues in embodied virtual agents for virtual environments
Embodied virtual agents (EVAs) are increasingly deployed in immersive virtual environments, where their nonverbal cues serve as a primary channel for affective communication. However, systematic evidence on how these nonverbal cues shape users’ emotional perception in virtual environments remains limited and dispersed across the literature. This paper presents a systematic literature review from a human-perception perspective to examine EVAs’ nonverbal cues in virtual environments and their effects on users’ affective perception. Searches were conducted in Scopus, the ACM Digital Library, IEEE Xplore, Web of Science, and SpringerLink, yielding 1066 records. Following the PRISMA guidelines, 52 studies met the inclusion criteria. The results indicate that facial expressions, posture and gaze are the most commonly used cues that influence users’ affective outcomes, with postural cues often demonstrating greater effectiveness in virtual environments. EVAs’ nonverbal expressions shape users’ affective perception across multiple dimensions, including their emotional responses, perceived social affect, and sense of social presence. More importantly, the immersive properties of virtual environments amplify the emotional contagion effects of these cues, while also increasing users’ sensitivity to subtle inconsistencies. Furthermore, the effectiveness of EVAs’ nonverbal cues in shaping users’ affective perceptions is context-dependent, with differing priorities across application domains. Advances in nonverbal generation for EVAs, from rule-based systems to generative AI, have progressively improved fidelity and affective perception, although they may also give rise to the uncanny valley effect. This review summarizes that EVAs’ nonverbal emotional cues substantially shape users’ affective perception in virtual environments, underscoring the importance of coherent multimodal cue design and context-appropriate generation methods for creating emotionally effective virtual agents.
- Open Access
Boundary Conditions for LLM-Generated Feedback in Primary Writing: An Educator-Aligned Evaluation and Design Considerations
Generative large language models (LLMs) are increasingly used to support writing feedback. However, the pedagogical safety and usefulness of LLM feedback for primary students remains under-evaluated. This study reports an educator-centered evaluation of GPT-4 Turbo for Year 5 narrative and persuasive writing in the context of an established online tutoring program. Using authentic students’ drafts paired with tutor feedback, we generated parallel LLM feedback via rubric-aligned prompting and compared the two feedback sources in a blinded, within-script design. Four experienced English specialists co-designed a six-dimensional rubric (clarity, specificity, helpfulness, feasibility, relevance, and overall effectiveness) and rated tutor versus LLM feedback for each script; their written reflections were analyzed thematically to surface boundary conditions and risk perceptions. Across dimensions, tutor feedback received slightly higher mean ratings, with the clearest descriptive advantage in perceived helpfulness; however, none of the differences remained statistically significant after Holm-Bonferroni correction. LLM feedback was often rated similarly for clarity and feasibility but was frequently characterized as generic, surface-focused, and occasionally misaligned with the student draft, which increased verification effort and posed a risk of misleading learners if used without mediation. Synthesizing ratings and educator reflections, we identify conditions under which LLM feedback is most appropriate as rapid first-pass support for routine structure and surface revision, and least appropriate for developmental judgment and context-sensitive guidance. We translate these findings into design requirements for teacher-in-the-loop primary writing feedback systems, including alignment to explicit pedagogical constructs, editable workflows, and safeguards that reduce unsupported feedback before release to students.
Open AccessBoundary Conditions for LLM-Generated Feedback in Primary Writing: An Educator-Aligned Evaluation and Design Considerations
Generative large language models (LLMs) are increasingly used to support writing feedback. However, the pedagogical safety and usefulness of LLM feedback for primary students remains under-evaluated. This study reports an educator-centered evaluation of GPT-4 Turbo for Year 5 narrative and persuasive writing in the context of an established online tutoring program. Using authentic students’ drafts paired with tutor feedback, we generated parallel LLM feedback via rubric-aligned prompting and compared the two feedback sources in a blinded, within-script design. Four experienced English specialists co-designed a six-dimensional rubric (clarity, specificity, helpfulness, feasibility, relevance, and overall effectiveness) and rated tutor versus LLM feedback for each script; their written reflections were analyzed thematically to surface boundary conditions and risk perceptions. Across dimensions, tutor feedback received slightly higher mean ratings, with the clearest descriptive advantage in perceived helpfulness; however, none of the differences remained statistically significant after Holm-Bonferroni correction. LLM feedback was often rated similarly for clarity and feasibility but was frequently characterized as generic, surface-focused, and occasionally misaligned with the student draft, which increased verification effort and posed a risk of misleading learners if used without mediation. Synthesizing ratings and educator reflections, we identify conditions under which LLM feedback is most appropriate as rapid first-pass support for routine structure and surface revision, and least appropriate for developmental judgment and context-sensitive guidance. We translate these findings into design requirements for teacher-in-the-loop primary writing feedback systems, including alignment to explicit pedagogical constructs, editable workflows, and safeguards that reduce unsupported feedback before release to students.
- Open Access
- 25 May 2026
- Published by MDPI AG in Preprints.org
Boundary Conditions for LLM-Generated Feedback in Primary Writing: An Educator-Aligned Evaluation and Design Considerations
Generative large language models (LLMs) are increasingly used to support writing feedback. However, the pedagogical safety and usefulness of LLM feedback for primary students remains under-evaluated. This study reports an educator-centered evaluation of GPT-4 Turbo for Year 5 narrative and persuasive writing in the context of an established online tutoring program. Using authentic students’ drafts paired with tutor feedback, we generated parallel LLM feedback via rubric-aligned prompting and compared the two feedback sources in a blinded, within-script design. Four experienced English specialists co-designed a six-dimensional rubric (clarity, specificity, helpfulness, feasibility, relevance, and overall effectiveness) and rated tutor versus LLM feedback for each script; their written reflections were analyzed thematically to surface boundary conditions and risk perceptions. Across dimensions, tutor feedback was rated higher overall, with the clearest advantage in perceived helpfulness and specificity. LLM feedback was often rated similarly for clarity and feasibility but was frequently characterized as generic, surface-focused, and occasionally misaligned with the student draft, which increased verification effort and posed a risk of misleading learners if used without mediation. Synthesizing ratings and educator reflections, we identify conditions under which LLM feedback is most appropriate as rapid first-pass support for routine structure and surface revision, and least appropriate for developmental judgment and context-sensitive guidance. We translate these findings into design requirements for teacher-in-the-loop primary writing feedback systems, including alignment to explicit pedagogical constructs, editable workflows, and safeguards that reduce unsupported feedback before release to students.
- 25 May 2026
- Published by MDPI AG in Preprints.org
Open AccessBoundary Conditions for LLM-Generated Feedback in Primary Writing: An Educator-Aligned Evaluation and Design Considerations
Generative large language models (LLMs) are increasingly used to support writing feedback. However, the pedagogical safety and usefulness of LLM feedback for primary students remains under-evaluated. This study reports an educator-centered evaluation of GPT-4 Turbo for Year 5 narrative and persuasive writing in the context of an established online tutoring program. Using authentic students’ drafts paired with tutor feedback, we generated parallel LLM feedback via rubric-aligned prompting and compared the two feedback sources in a blinded, within-script design. Four experienced English specialists co-designed a six-dimensional rubric (clarity, specificity, helpfulness, feasibility, relevance, and overall effectiveness) and rated tutor versus LLM feedback for each script; their written reflections were analyzed thematically to surface boundary conditions and risk perceptions. Across dimensions, tutor feedback was rated higher overall, with the clearest advantage in perceived helpfulness and specificity. LLM feedback was often rated similarly for clarity and feasibility but was frequently characterized as generic, surface-focused, and occasionally misaligned with the student draft, which increased verification effort and posed a risk of misleading learners if used without mediation. Synthesizing ratings and educator reflections, we identify conditions under which LLM feedback is most appropriate as rapid first-pass support for routine structure and surface revision, and least appropriate for developmental judgment and context-sensitive guidance. We translate these findings into design requirements for teacher-in-the-loop primary writing feedback systems, including alignment to explicit pedagogical constructs, editable workflows, and safeguards that reduce unsupported feedback before release to students.
- Open Access
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-19
Exploring Design Practice with Generative AI: Perspectives from AEC Design Professionals
The integration of Generative AI (GenAI) into Architecture, Engineering, and Construction (AEC) design marks a paradigm shift, yet empirical studies on AEC professional adoption are scarce. This study aims to bridge this gap through a mixed-methods study involving interviews with 20 professionals and a survey of 191 professionals. Findings show GenAI is mainly used in conceptual design, improving client and intra-team collaboration, especially in concept-only projects, strong site integration projects, and challenging client engagements projects for inspiration exploration, client communication and design service boundaries expansion. We further uncover a critical paradox: professionals perceive GenAI’s creativity as intrinsically linked to the unpredictability. Additionally, a nuanced “shame-yet-pride” dynamic is identified, wherein professionals publicly regard GenAI as a key asset but conceal its use to exaggerate manual effort. Based on these insights, we recommend fostering an industry-wide discussion for a mindset shift toward transparent collaboration with GenAI. This study offers the first comprehensive empirical study for understanding GenAI’s current role and future potential in transforming AEC design practices.
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-19
Open AccessExploring Design Practice with Generative AI: Perspectives from AEC Design Professionals
The integration of Generative AI (GenAI) into Architecture, Engineering, and Construction (AEC) design marks a paradigm shift, yet empirical studies on AEC professional adoption are scarce. This study aims to bridge this gap through a mixed-methods study involving interviews with 20 professionals and a survey of 191 professionals. Findings show GenAI is mainly used in conceptual design, improving client and intra-team collaboration, especially in concept-only projects, strong site integration projects, and challenging client engagements projects for inspiration exploration, client communication and design service boundaries expansion. We further uncover a critical paradox: professionals perceive GenAI’s creativity as intrinsically linked to the unpredictability. Additionally, a nuanced “shame-yet-pride” dynamic is identified, wherein professionals publicly regard GenAI as a key asset but conceal its use to exaggerate manual effort. Based on these insights, we recommend fostering an industry-wide discussion for a mindset shift toward transparent collaboration with GenAI. This study offers the first comprehensive empirical study for understanding GenAI’s current role and future potential in transforming AEC design practices.
- Open Access
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-6
Auto-Generating Personas from User Reviews in VR App Stores
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-6
Open AccessAuto-Generating Personas from User Reviews in VR App Stores
- Open Access
- 25 April 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-17
Understanding VR Accessibility Practices of VR Professionals
- 25 April 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-17
Open AccessUnderstanding VR Accessibility Practices of VR Professionals
- Open Access
- 9 November 2024
- Published by Elsevier in Information and Software Technology
- Vol. 178
Who uses personas in requirements engineering: The practitioners’ perspective
- 9 November 2024
- Published by Elsevier in Information and Software Technology
- Vol. 178
Open AccessWho uses personas in requirements engineering: The practitioners’ perspective
- Open Access
- 11 May 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Auto-Generated Personas: Enhancing User-centered Design Practices among University Students
Personas are commonly used in User-centered Design (UCD) activities to help designers better understand users’ needs. However, there is still a reliance on traditional approaches such as interviews and ethnography for building personas in UCD activities. To this end, we developed an auto-generating persona system to enhance practices in UCD course activities. Our persona system is developed based on the GPT-4 model, the DALL-E 2 model, and knowledge graphs. Hence, our persona system includes three main features of our persona system: automated processing of survey data, automatic generation of 2D avatars, and providing options for automatic or customized entity generation. We recruited a total of 22 participants to evaluate our persona system. Our findings confirmed that there was a significant difference in terms of efficiency, satisfaction, accuracy, and diversity. Meanwhile, participants provided both positive and negative feedback regarding our persona system. As on-going work, we discuss the current limitations of our persona system and explore future research directions to further improve its capabilities and effectiveness.
- 11 May 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessAuto-Generated Personas: Enhancing User-centered Design Practices among University Students
Personas are commonly used in User-centered Design (UCD) activities to help designers better understand users’ needs. However, there is still a reliance on traditional approaches such as interviews and ethnography for building personas in UCD activities. To this end, we developed an auto-generating persona system to enhance practices in UCD course activities. Our persona system is developed based on the GPT-4 model, the DALL-E 2 model, and knowledge graphs. Hence, our persona system includes three main features of our persona system: automated processing of survey data, automatic generation of 2D avatars, and providing options for automatic or customized entity generation. We recruited a total of 22 participants to evaluate our persona system. Our findings confirmed that there was a significant difference in terms of efficiency, satisfaction, accuracy, and diversity. Meanwhile, participants provided both positive and negative feedback regarding our persona system. As on-going work, we discuss the current limitations of our persona system and explore future research directions to further improve its capabilities and effectiveness.