Feng Ji
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
Feng Ji has a total of 117 co-authors. Here are the top co-authors by number of shared publications:
- Jinbo He
Department of Biosciences and Bioinformatics, School of Science, Xi’an Jiaotong-Liverpool University, Suzhou, China
19 shared publications - Wesley R. Barnhart
Department of Psychology Suffolk University Boston Massachusetts USA
11 shared publications - Yuting Han
Beijing Language and Culture University
9 shared publications - Nanyu Luo
Department of Applied Psychology and Human Development, University of Toronto, Toronto, Canada
6 shared publications - Xiaoya Zhang
Department of Family, Youth and Community Sciences, Institute of Food and Agricultural Sciences, University of Florida, Gainesville, USA
5 shared publications - Yuchen Zhang
Department of Applied Psychology and Human Development, University of Toronto, Toronto, Canada
5 shared publications
Latest Publications
- Open Access
- 28 October 2026
- Published by Springer Nature in Nature Communications
The integrated environmental zoning-based regulation further enhances sustainable development in China
- 28 October 2026
- Published by Springer Nature in Nature Communications
Open AccessThe integrated environmental zoning-based regulation further enhances sustainable development in China
- Open Access
- 17 September 2026
- Published by Springer Nature in Journal of Eating Disorders
Identifying and characterizing eating disorder discourse on Chinese social media: a machine learning approach
- 17 September 2026
- Published by Springer Nature in Journal of Eating Disorders
Open AccessIdentifying and characterizing eating disorder discourse on Chinese social media: a machine learning approach
- 16 September 2026
- Published by SAGE Publications in Journal of Educational Computing Research
Measuring Secondary School Students’ Verification Capability in GenAI-Supported Learning: Development and Application of the α − v − M Framework
Generative Artificial Intelligence (GenAI) provides adolescents with fluent and immediate support, but unverified use may encourage cognitive offloading and passive reliance. This study conceptualizes verification capability as a cognitive defense mechanism through which learners inspect, question, correct, compare, or reconstruct GenAI outputs before accepting them. Using an exploratory sequential mixed-methods design, we first conducted deductive qualitative coding of 762 authentic Human–AI collaborative teaching cases. Among 242 valid GenAI-supported cases, 98.8% showed no documented verification of AI-generated outputs, indicating a severe observable verification deficit. We then developed and validated the α − v − M framework, comprising AI Engagement, Verification Intensity, and Multi-model Cross-validation, through a scale study with 422 secondary school students. Psychometric validation provided generally supportive evidence for the scale structure, and Item Response Theory was used to evaluate item functioning and generate latent trait scores. Structural equation modeling showed that AI engagement was associated with human–AI collaborative quality through verification-related processes. Gaussian graphical modeling identified deep logical auditing as a central cognitive defense indicator. Latent profile analysis further revealed four learner profiles: Naïve Trusters, Superficial Checkers, Isolated Auditors, and Symbiotic Strategists. These findings inform GenAI-supported learning design by emphasizing cognitive friction, verification scaffolds, and profile-sensitive feedback.
- 16 September 2026
- Published by SAGE Publications in Journal of Educational Computing Research
Measuring Secondary School Students’ Verification Capability in GenAI-Supported Learning: Development and Application of the α − v − M Framework
Generative Artificial Intelligence (GenAI) provides adolescents with fluent and immediate support, but unverified use may encourage cognitive offloading and passive reliance. This study conceptualizes verification capability as a cognitive defense mechanism through which learners inspect, question, correct, compare, or reconstruct GenAI outputs before accepting them. Using an exploratory sequential mixed-methods design, we first conducted deductive qualitative coding of 762 authentic Human–AI collaborative teaching cases. Among 242 valid GenAI-supported cases, 98.8% showed no documented verification of AI-generated outputs, indicating a severe observable verification deficit. We then developed and validated the α − v − M framework, comprising AI Engagement, Verification Intensity, and Multi-model Cross-validation, through a scale study with 422 secondary school students. Psychometric validation provided generally supportive evidence for the scale structure, and Item Response Theory was used to evaluate item functioning and generate latent trait scores. Structural equation modeling showed that AI engagement was associated with human–AI collaborative quality through verification-related processes. Gaussian graphical modeling identified deep logical auditing as a central cognitive defense indicator. Latent profile analysis further revealed four learner profiles: Naïve Trusters, Superficial Checkers, Isolated Auditors, and Symbiotic Strategists. These findings inform GenAI-supported learning design by emphasizing cognitive friction, verification scaffolds, and profile-sensitive feedback.
- Open Access
- 9 September 2026
- Published by International Society for Data Science and Analytics in Journal of Behavioral Data Science
- Vol. 6 (2) , 173-184
Can Large Language Models (LLMs) be Trusted for Power Analysis? An Empirical Evaluation
Power analysis is critical for assuring rigor and validity of quantitative research yet remains underutilized due to technical challenges associated with specialized software. At the same time, large language models (LLMs) are being rapidly integrated into research practice, raising interest in their potential to assist statistical and research design tasks. However, despite their widespread adoption, the reliability of LLMs in supporting statistically rigorous procedures has not been systematically evaluated, posing risks for unexamined or overly optimistic use. To address this gap, we evaluated four widely used LLMs—ChatGPT (GPT-3.5, GPT-4, GPT-4o) and Llama 3.2—across two experiments. Experiment 1 examined whether LLMs could calculate required sample sizes for common statistical tests (two-sample t-test, one-way ANOVA, and χ² goodness-of-fit test) under different prompting strategies, including direct calculation versus R/Python code generation. Experiment 2 assessed models’ ability to identify missing input parameters necessary for power analysis, which is a task that requires methodological understanding. Results revealed that GPT-4 and GPT-4o performed well when generating R code for sample size estimation, but struggled with direct numerical calculation. Furthermore, while LLMs were able to detect missing information, their reliability varied by statistical context. Findings suggest that while LLMs may offer support in structuring and initiating power analysis, they cannot substitute for expert judgment. Overall, the study underscores the importance of critically evaluating LLM performance in statistically demanding tasks. Responsible integration of LLM requires critical oversight, cross-verification, and methodological evaluation.
- 9 September 2026
- Published by International Society for Data Science and Analytics in Journal of Behavioral Data Science
- Vol. 6 (2) , 173-184
Open AccessCan Large Language Models (LLMs) be Trusted for Power Analysis? An Empirical Evaluation
Power analysis is critical for assuring rigor and validity of quantitative research yet remains underutilized due to technical challenges associated with specialized software. At the same time, large language models (LLMs) are being rapidly integrated into research practice, raising interest in their potential to assist statistical and research design tasks. However, despite their widespread adoption, the reliability of LLMs in supporting statistically rigorous procedures has not been systematically evaluated, posing risks for unexamined or overly optimistic use. To address this gap, we evaluated four widely used LLMs—ChatGPT (GPT-3.5, GPT-4, GPT-4o) and Llama 3.2—across two experiments. Experiment 1 examined whether LLMs could calculate required sample sizes for common statistical tests (two-sample t-test, one-way ANOVA, and χ² goodness-of-fit test) under different prompting strategies, including direct calculation versus R/Python code generation. Experiment 2 assessed models’ ability to identify missing input parameters necessary for power analysis, which is a task that requires methodological understanding. Results revealed that GPT-4 and GPT-4o performed well when generating R code for sample size estimation, but struggled with direct numerical calculation. Furthermore, while LLMs were able to detect missing information, their reliability varied by statistical context. Findings suggest that while LLMs may offer support in structuring and initiating power analysis, they cannot substitute for expert judgment. Overall, the study underscores the importance of critically evaluating LLM performance in statistically demanding tasks. Responsible integration of LLM requires critical oversight, cross-verification, and methodological evaluation.
- Open Access
- 19 August 2026
- Published by Center for Open Science in PsyArXiv
The Generative Artificial Intelligence Literacy Scale (GAILS): Development, Validation, and Measurement Invariance Across Sex and Occupational Status Groups
Generative Artificial Intelligence (GenAI) tools are rapidly reshaping how people seek information, learn, and work, while also raising concerns about misinformation, bias, privacy, and academic integrity across both educational and workplace contexts. However, many existing AI literacy scales targeted broad AI knowledge rather than GenAI-specific competencies (e.g., prompt formulation, evaluating generated outputs). Emerging GenAI literacy measures have also focused primarily on university students and non-North American samples, limiting their generalizability. We developed and validated a GenAI literacy scale (GAILS) in a diverse North American adult sample (N = 341; Mage = 39.00 years; 38.42% students, 61.58% workforce). The GAILS showed excellent internal consistency (α = 0.973, 95% CI [0.968, 0.978]). Exploratory factor analysis supported a three-factor solution (KMO = 0.959; Bartlett’s χ²(561) = 5784.58, p < .001), and confirmatory factor analysis further supported the model (CFI = 0.967, TLI = 0.965, SRMR = 0.058, RMSEA = 0.087). Items were designed to assess three domains: (a) Adaptive Operational Skills, (b) Responsible GenAI Literacy, and (c) Critical Evaluation & Autonomous Use. Measurement invariance analyses supported scalar invariance across sex (female vs. male) and occupational status (student vs. workforce) groups. Construct validity was supported by positive correlations with GenAI acceptance (r = 0.81) and GenAI trust (r = 0.47). Semantic embedding analysis further supported the conceptual structure of the scale by showing distinct semantic clustering for Responsible GenAI Literacy and meaningful conceptual proximity between Adaptive Operational Skills and Critical Evaluation & Autonomous Use. These results suggest that the GAILS is a psychometrically and conceptually supported tool for assessing GenAI literacy gaps and informing educational, workplace, and policy interventions.
- 19 August 2026
- Published by Center for Open Science in PsyArXiv
Open AccessThe Generative Artificial Intelligence Literacy Scale (GAILS): Development, Validation, and Measurement Invariance Across Sex and Occupational Status Groups
Generative Artificial Intelligence (GenAI) tools are rapidly reshaping how people seek information, learn, and work, while also raising concerns about misinformation, bias, privacy, and academic integrity across both educational and workplace contexts. However, many existing AI literacy scales targeted broad AI knowledge rather than GenAI-specific competencies (e.g., prompt formulation, evaluating generated outputs). Emerging GenAI literacy measures have also focused primarily on university students and non-North American samples, limiting their generalizability. We developed and validated a GenAI literacy scale (GAILS) in a diverse North American adult sample (N = 341; Mage = 39.00 years; 38.42% students, 61.58% workforce). The GAILS showed excellent internal consistency (α = 0.973, 95% CI [0.968, 0.978]). Exploratory factor analysis supported a three-factor solution (KMO = 0.959; Bartlett’s χ²(561) = 5784.58, p < .001), and confirmatory factor analysis further supported the model (CFI = 0.967, TLI = 0.965, SRMR = 0.058, RMSEA = 0.087). Items were designed to assess three domains: (a) Adaptive Operational Skills, (b) Responsible GenAI Literacy, and (c) Critical Evaluation & Autonomous Use. Measurement invariance analyses supported scalar invariance across sex (female vs. male) and occupational status (student vs. workforce) groups. Construct validity was supported by positive correlations with GenAI acceptance (r = 0.81) and GenAI trust (r = 0.47). Semantic embedding analysis further supported the conceptual structure of the scale by showing distinct semantic clustering for Responsible GenAI Literacy and meaningful conceptual proximity between Adaptive Operational Skills and Critical Evaluation & Autonomous Use. These results suggest that the GAILS is a psychometrically and conceptually supported tool for assessing GenAI literacy gaps and informing educational, workplace, and policy interventions.
- Open Access
- 23 July 2026
- Published by SAGE Publications in Applied Psychological Measurement
avbirt: A Python Package for Flexible Item Factor Analysis with Adversarial Variational Bayes
- 23 July 2026
- Published by SAGE Publications in Applied Psychological Measurement
Open Accessavbirt: A Python Package for Flexible Item Factor Analysis with Adversarial Variational Bayes
- 14 July 2026
- Published by Elsevier in Body Image
- Vol. 58, 102149
Adverse childhood experiences and body appreciation: The buffering role of posttraumatic growth
- 14 July 2026
- Published by Elsevier in Body Image
- Vol. 58, 102149
Adverse childhood experiences and body appreciation: The buffering role of posttraumatic growth
- Open Access
- 5 July 2026
- Published by Center for Open Science in PsyArXiv
Clarifying the Conceptual Landscape in AI Literacy Measurement: A Large Language Model Based Approach
The rapid proliferation of AI literacy instruments has expanded the scope of the field, while introducing diversity in definitions, dimensional structures, and item operationalizations, which complicates cross-instrument comparison and cumulative interpretation. To support a clearer view of this measurement landscape, this study employs a large language model (LLM) based semantic analysis to map the structure of AI literacy content domain across instruments. Drawing on 11 rigorously reviewed AI literacy scales comprising 55 constructs and 272 items, we map construct definitions and item texts into a shared semantic space using transformer-based embeddings. The analysis first examines the coherence and distinctiveness of AI literacy constructs across instruments, and then explores patterns of alignment and divergence between construct labels and their item operationalizations. The results indicate a semantically consolidated core centered on cognitive and technical dimensions of AI literacy, alongside more weakly integrated affective and collaborative components. In addition, patterns of semantic convergence and divergence are observed. Constructs with different labels (e.g., Behavioral Commitment and AI Learning) show similarity in their item operationalizations, whereas other constructs sharing similar labels (e.g., Technical Proficiency and Technical Understanding) diverge in the competencies emphasized by their items. Methodologically, this filed-level mapping provides a foundation for more rigorous instrument development and supports cumulative measurement development. Our results highlight the importance of using clearer and more internally consistent operationalizations to support cumulative progress in measurement development and to enhance the educational applicability of AI literacy frameworks.
- 5 July 2026
- Published by Center for Open Science in PsyArXiv
Open AccessClarifying the Conceptual Landscape in AI Literacy Measurement: A Large Language Model Based Approach
The rapid proliferation of AI literacy instruments has expanded the scope of the field, while introducing diversity in definitions, dimensional structures, and item operationalizations, which complicates cross-instrument comparison and cumulative interpretation. To support a clearer view of this measurement landscape, this study employs a large language model (LLM) based semantic analysis to map the structure of AI literacy content domain across instruments. Drawing on 11 rigorously reviewed AI literacy scales comprising 55 constructs and 272 items, we map construct definitions and item texts into a shared semantic space using transformer-based embeddings. The analysis first examines the coherence and distinctiveness of AI literacy constructs across instruments, and then explores patterns of alignment and divergence between construct labels and their item operationalizations. The results indicate a semantically consolidated core centered on cognitive and technical dimensions of AI literacy, alongside more weakly integrated affective and collaborative components. In addition, patterns of semantic convergence and divergence are observed. Constructs with different labels (e.g., Behavioral Commitment and AI Learning) show similarity in their item operationalizations, whereas other constructs sharing similar labels (e.g., Technical Proficiency and Technical Understanding) diverge in the competencies emphasized by their items. Methodologically, this filed-level mapping provides a foundation for more rigorous instrument development and supports cumulative measurement development. Our results highlight the importance of using clearer and more internally consistent operationalizations to support cumulative progress in measurement development and to enhance the educational applicability of AI literacy frameworks.
- 28 June 2026
- Published by Springer Nature
- p. 361-369
A Generator-Aligner Pipeline for LLM-Based Situational Judgment Test Generation
- 28 June 2026
- Published by Springer Nature
- p. 361-369
A Generator-Aligner Pipeline for LLM-Based Situational Judgment Test Generation
- Open Access
- 26 June 2026
- Published by MDPI AG in Behavioral Sciences
- Vol. 16 (7) , 1064
Multi-Agent LLMs for Occupational Profiling: Psychometric Validation on 1636 Chinese Occupations
Occupation-level psychological profiles, such as RIASEC interests and Big Five personality, underpin career counseling, person–job matching, and workforce research, but building them at scale has been expensive and limited to a few national taxonomies. The O*NET Interest Profiler, the largest operationalization of RIASEC, took more than two decades of worker surveys, expert ratings, and iterative empirical calibration to construct, and the 2022 Chinese Occupational Classification has no comparable psychological database. Large language models (LLMs) offer a scalable alternative, but using them as raters raises issues that single-model designs do not resolve: inter-rater reliability, calibration to external benchmarks, and systematic psychometric validation. We propose a multi-agent LLM framework in which three LLMs serve as separate expert raters, in-context anchors align the rating scale, and a separate arbitrator resolves rater disagreements. We applied the framework to all 1636 occupations in the 2022 Chinese Occupational Classification, producing six RIASEC and five Big Five scores per occupation. RIASEC dimensions showed uniformly excellent reliability (intraclass correlation coefficient, ICC [2,1] = 0.87 to 0.98) and high convergent correlations with O*NET (r = 0.84 to 0.96); structural validity received weak support (Tracey’s C = 0.653, ns), though the dimensions differentiated occupational categories as theory predicts, and the profile space recovered the administrative taxonomy (adjusted Rand index, ARI = 0.418). Big Five absolute agreement was uniformly high, although ICC(2,1) values for Conscientiousness and Neuroticism were attenuated by variance compression and model-level calibration offsets rather than rater disagreement. The Big Five scores are, therefore, suited to broad occupational differentiation, particularly on Openness and Extraversion, rather than to fine-grained rank ordering on Conscientiousness or Neuroticism. The framework also yields the first occupation-level RIASEC and Big Five database for the 2022 Chinese Occupational Classification, openly available for applied use.
- 26 June 2026
- Published by MDPI AG in Behavioral Sciences
- Vol. 16 (7) , 1064
Open AccessMulti-Agent LLMs for Occupational Profiling: Psychometric Validation on 1636 Chinese Occupations
Occupation-level psychological profiles, such as RIASEC interests and Big Five personality, underpin career counseling, person–job matching, and workforce research, but building them at scale has been expensive and limited to a few national taxonomies. The O*NET Interest Profiler, the largest operationalization of RIASEC, took more than two decades of worker surveys, expert ratings, and iterative empirical calibration to construct, and the 2022 Chinese Occupational Classification has no comparable psychological database. Large language models (LLMs) offer a scalable alternative, but using them as raters raises issues that single-model designs do not resolve: inter-rater reliability, calibration to external benchmarks, and systematic psychometric validation. We propose a multi-agent LLM framework in which three LLMs serve as separate expert raters, in-context anchors align the rating scale, and a separate arbitrator resolves rater disagreements. We applied the framework to all 1636 occupations in the 2022 Chinese Occupational Classification, producing six RIASEC and five Big Five scores per occupation. RIASEC dimensions showed uniformly excellent reliability (intraclass correlation coefficient, ICC [2,1] = 0.87 to 0.98) and high convergent correlations with O*NET (r = 0.84 to 0.96); structural validity received weak support (Tracey’s C = 0.653, ns), though the dimensions differentiated occupational categories as theory predicts, and the profile space recovered the administrative taxonomy (adjusted Rand index, ARI = 0.418). Big Five absolute agreement was uniformly high, although ICC(2,1) values for Conscientiousness and Neuroticism were attenuated by variance compression and model-level calibration offsets rather than rater disagreement. The Big Five scores are, therefore, suited to broad occupational differentiation, particularly on Openness and Extraversion, rather than to fine-grained rank ordering on Conscientiousness or Neuroticism. The framework also yields the first occupation-level RIASEC and Big Five database for the 2022 Chinese Occupational Classification, openly available for applied use.