CoChat
Sign Up
Cited by 9Technology Adoption and User BehaviourArtificial Intelligence in EducationArtificial Intelligence in Healthcare and Education

A two-step structural equation modeling and explainable machine learning framework for understanding university students’ adoption of generative AI: balancing intrinsic motivations and perceived risks

Open access
Zeng, Daixing, Xu, Xiaoqin, Zhu, Tianxing, Li, Yong, Li, Qiumin·2026·Frontiers in Psychology
DOI 10.3389/fpsyg.2026.1743722PMC PMC12894297
Open this paper in CoChat — free accountOpen PDF

Abstract

The adoption of generative AI tools by university students has surged, embodying a mix of promising benefits and serious concerns. Understanding the factors that drive or hinder students’ adoption of GenAI is essential for responsible integration of AI technologies in higher education. This study introduces a novel two step SEM–XML framework that couples structural equation modeling (SEM) with an explainable machine learning (XML) component, overcoming limitations of traditional SEM and enabling both hypothesis-driven path analysis and data-driven factor identification. Grounded in an integrated benefit–risk perspective, this framework blends constructs from the Technology Acceptance Model, Theory of Planned Behavior, Perceived Risk, and Knowledge Attitude Practice models, emphasizing students’ intrinsic motivations. The study is designed as a cross-sectional survey, with an effective sample size of 880 respondents from southwestern China, including undergraduate, master’s, and doctoral students. The average age of participants is 20.8 years, with a gender distribution of 48.52% male and a diverse academic background, encompassing fields such as Engineering, Economics, Science, and Management. We test this framework using a survey of university students’ GenAI usage. Results show that positive perceptions such as perceived usefulness and personal interest strongly encourage GenAI use. In contrast, perceived risks related to ethics, accuracy, and academic integrity significantly inhibit it. This pattern is partially consistent with previous findings on ChatGPT adoption. These findings highlight how internal attitudes and external pressures interact to shape GenAI uptake. This study emphasizes the substantial impact of both internal and external factors on students’ acceptance of GenAI tools, providing valuable insights for educational institutions, policymakers, and tool developers.

Verified 2026-10-10 · 1 reference not checked yet

AI snapshot

Written from the paper’s full text.

TL;DR

This study proposes a two-step SEM–XML framework that combines structural equation modeling with explainable machine learning to examine university students’ adoption of generative AI through an integrated benefit–risk perspective. In 880 valid survey responses from southwestern China, perceived usefulness, ease of use, subjective norms, and technology awareness promoted adoption intention/use, while perceived personal academic and academic environmental risks inhibited it; perceived social risk was not significant.

Key findings

  • SEM path results: perceived usefulness → behavioral intention (β = 0.544, p < 0.001); perceived ease of use → behavioral intention (β = 0.100, p < 0.05); behavioral intention → actual behavior (β = 0.824, p < 0.001); subjective norm → behavioral intention (β = 0.339, p < 0.001); technology awareness → behavioral intention (β = 0.166, p < 0.001).
  • Risk paths: perceived personal academic risk → behavioral intention (β = −0.316, p < 0.001); perceived academic environmental risk → behavioral intention (β = −0.094, p < 0.05); perceived social risk → behavioral intention was not significant (β = −0.008, p > 0.05) and H7 was rejected.
  • XML stage: XGBoost was selected as the final model with test R² = 0.9867 and cross-validation R² = 0.9812; top factors included TA2, TA1, PE1, PU1, PU2, PAR4, PER2, and PER3; 8 of the top 10 features overlapped across the two interpretability methods.
  • Descriptive results: 78% of respondents had used GenAI tools and 22% had not; among non-users, 34.5% expressed willingness to use GenAI tools in the future.

Methods

  • Cross-sectional survey conducted in Southwestern China from March to April 2025 using two-stage sampling; 900 questionnaires were distributed (420 offline, 480 online), yielding 880 valid responses and an effective response rate of 97.78%; the sample included undergraduate, master’s, and doctoral students, with the abstract reporting an average age of 20.8 years and 48.52% male.
  • Questionnaire included three demographic questions and nine multi-item constructs measured on a five-point Likert scale: Technology Awareness (TA), Perceived Ease of Use (PE), Perceived Usefulness (PU), Perceived Personal Academic Risk (PAR), Perceived Academic Environmental Risk (PER), Perceived Social Risk (PSR), Subjective Norm (SN), Behavioral Intention (BI), and Actual Behavior (AB); a pilot test was conducted with 15 participants.
  • Analysis proceeded in two steps: Step 1 used SEM in AMOS 26.0 with measurement model/CFA and path tests H1–H8; Step 2 used explainable machine learning with eight algorithms (Linear Regression, Lasso, Ridge, KNN, SVR, Random Forest, XGBoost, Gradient Boosting), a 70:30 train-test split, 10-fold cross-validation, grid search, RMSE optimization, and R²; XGBoost feature importance was assessed via gain and SHAP, with core factors identified only when both methods agreed.

Limitations

  • The paper states that the sample is limited to university students from Southwest China, restricting generalizability; self-reported data may be subject to response bias; the cross-sectional design limits causal inferences; and space constraints limited comprehensiveness, leaving subgroup or moderation analysis for future work.
  • The text makes evident that the XML stage does not explicitly propagate SEM measurement error or uncertainty; Step 2 is informed by the SEM measurement structure and is not an independent corroboration of SEM path significance; the authors also acknowledge possible overfitting in XGBoost due to the deep tree structure and lack of subsampling.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3389/fpsyg.2026.1743722

Cites

77 of 78 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 28 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Positive perceptions such as perceived usefulness and personal interest strongly encourage university students' GenAI use.

    “Results show that positive perceptions such as perceived usefulness and personal interest strongly encourage GenAI use.”
    Quote in the paperSubstantial claim
  • Findingown data

    Perceived risks related to ethics, accuracy, and academic integrity significantly inhibit university students' GenAI use.

    “In contrast, perceived risks related to ethics, accuracy, and academic integrity significantly inhibit it.”
    Quote in the paperSubstantial claim
  • Findingown data

    The relationship between education level and perceived benefits and risks was non-linear.

    “The relationship with education level was non-linear, with master’s students reporting the highest perceived benefits and risks, while PhD students had significantly lower perceived benefits than both undergraduates and master’s students.”
    Quote in the paperNot a substantial claim
  • Findingown data

    PhD students had significantly lower perceived benefits than both undergraduates and master's students.

    “The relationship with education level was non-linear, with master’s students reporting the highest perceived benefits and risks, while PhD students had significantly lower perceived benefits than both undergraduates and master’s students.”
    Quote in the paperNot a substantial claim
  • Findingown data

    Significant inter-group differences were found based on GenAI tool usage.

    “As shown in Table 3, significant inter-group differences were found based on GenAI tool usage.”
    Quote in the paperNot a substantial claim
  • May combine two claims. 37
  • Classified as probably not a substantial claim. 55
  • The quote only partly states this claim. 1

  • May combine two claims.

    Findingown data

    The group that used GenAI tools reported significantly higher perceived benefits and perceived risks than the group that had not used them.

    “The group that used GenAI tools reported significantly higher perceived benefits and perceived risks.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    University students mainly use GenAI tools for mastering theoretical knowledge (79.21%), building a knowledge framework (59.74%), and quickly grasping interdisciplinary knowledge (55.20%).

    “University students mainly use GenAI tools for master theoretical knowledge (e.g., answering individual concepts or basic issues) (79.21%), building a knowledge framework (e.g., understanding the distinctions and logical connections between related concepts) (59.74%), and quickly grasp interdisciplinary knowledge (55.20%).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Among the surveyed university students, 78% have used GenAI tools and 22% have not.

    “Among the respondents, 78% of university students have used GenAI tools, while only 22% have not.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingreasoning

    Feedback from the pilot helped refine the wording and structure of the items.

    “Feedback from the pilot helped refine the wording and structure of the items.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    TA2 is the most crucial influencing factor, significantly driving students' intention to adopt GenAI tools.

    “The findings show that TA2 (understanding of deep learning, large models, and related technical principles) is the most crucial influencing factor, significantly driving students’ intention to adopt GenAI tools.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    PE1, PU1, and PU2 were ranked among the top 10 in both methods.

    “PE1 (ease of use and accessibility), PE1 (helping to master theoretical knowledge), and PUT (supporting the establishment of knowledge frameworks) were all ranked among the top 10 in both methods”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    XGBoost demonstrated superior predictive accuracy and convergence (test R2 = 0.9867; cross-validation R2 = 0.9812), outperforming other algorithms.

    “Upon evaluation, XGBoost demonstrated superior predictive accuracy and convergence (test R2 = 0.9867; cross-validation R2 = 0.9812), outperforming other algorithms (see Figure 4).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Other core factors include TA1, PE1, PU1, PU2, and risks like PAR4, PER2, PER3.

    “Other core factors include TA1 (ability to distinguish between AI, AGI, and GenAI concepts), PE1 (ease of use), PU1 (helps in mastering theoretical knowledge), PU2 (assists in building knowledge frameworks), and risks like PAR4 (false data risk), PER2 (failure of plagiarism detection systems), and PER3 (academic result duplication).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    PAR4, PER2, and PER3 ranked highly.

    “At the same time, features such as PAR4 (risk of false data), PER2 (failure of plagiarism detection systems), and PER3 (duplication of academic results) ranked highly”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    GenAI technologies pose challenges such as undermining academic integrity and reducing the effectiveness of plagiarism detection systems.

    “Although GenAI technologies have transformed academic practices, they simultaneously pose challenges, such as undermining academic integrity and reducing the effectiveness of plagiarism detection systems (Huang et al., 2025).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Huang et al., 2025

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Students may perceive personal academic risks such as exposure to misleading information, reduced critical or creative thinking, weakened expression and analytical ability, or the use of inaccurate or fabricated data.

    “Students may perceive personal academic risks such as exposure to misleading information, reduced critical or creative thinking, weakened expression and analytical ability, or the use of inaccurate or fabricated data (Zhang L. et al., 2025; Zhang X. et al., 2025; Huang and Wu, 2025; Chan and Lee, 2025).”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites Zhang L. et al., 2025

      Not checked in the free run.

    • cites Zhang X. et al., 2025

      Not checked in the free run.

    • cites Huang and Wu, 2025

      Not checked in the free run.

  • May combine two claims.

    Interpretationnone given

    The study introduces a novel two-step SEM-XML framework that couples structural equation modeling (SEM) with an explainable machine learning (XML) component, overcoming limitations of traditional SEM and enabling both hypothesis-driven path analysis and data-driven factor identification.

    “This study introduces a novel two step SEM–XML framework that couples structural equation modeling (SEM) with an explainable machine learning (XML) component, overcoming limitations of traditional SEM and enabling both hypothesis-driven path analysis and data-driven factor identification.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: SEM is typically used to analyze direct, indirect, and moderating relationships between latent variables, but it lacks a deeper examination of measured variables.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    The more positive one's attitude, the stronger the perceived social norms and behavioral control, and the stronger the behavioral intention-ultimately increasing the likelihood of performing the behavior.

    “The more positive one’s attitude, the stronger the perceived social norms and behavioral control, and the stronger the behavioral intention—ultimately increasing the likelihood of performing the behavior (Ajzen, 1991).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Ajzen, 1991 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Perceived risk consists of both uncertainty and consequence severity.

    “Its core idea is that consumers’ decision-making is accompanied by subjective perceptions of uncertainty regarding outcomes, and perceived risk consists of both uncertainty and consequence severity (Barach, 1969).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Barach, 1969 source

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    The fundamental assumption of the Theory of Reasoned Action is that users' beliefs influence their behavioral intentions, and these intentions, in turn, determine their actual behavior toward using a given technology.

    “The fundamental assumption of the theory is that users’ beliefs influence their behavioral intentions, and these intentions, in turn, determine their actual behavior toward using a given technology (Taylor and Todd, 1995).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Taylor and Todd, 1995 source

      Not checked in the free run.

  • May combine two claims.

    Premisenone given

    Few studies have integrated benefits, risks, and cognitive theories to examine the adoption of GenAI in higher education, and even fewer have concretized the influencing factors through post-path analysis.

    “Few studies have integrated benefits, risks, and cognitive theories to examine the adoption of GenAI in higher education, and even fewer have concretized the influencing factors through post-path analysis.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The study introduces the 2SSX (two-step SEM-XML) research framework.

    “To address these limitations, this study introduces the “2SSX” (two-step SEM–XML) research framework.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: SEM is typically used to analyze direct, indirect, and moderating relationships between latent variables, but it lacks a deeper examination of measured variables.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    TAM identifies two core determinants of technology adoption intention: perceived usefulness and perceived ease of use, which jointly determine users' behavioral intentions.

    “It identifies two core determinants of technology adoption intention: perceived usefulness (PU) and perceived ease of use (PE) which jointly determine users’ behavioral intentions.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Davis, 1989 source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The dual-benefit-risk theoretical model of GenAI adoption integrates the Technology Acceptance Model, the Theory of Planned Behavior, Perceived Risk Theory, and the Knowledge-Attitude-Practice Model.

    “Developing a dual-benefit–risk theoretical model of GenAI adoption based on a synthesis of existing studies. This model integrates key frameworks such as the Technology Acceptance Model, the Theory of Planned Behavior, Perceived Risk Theory, and the Knowledge-Attitude-Practice Model.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    In the second stage, explainable machine learning techniques are used to identify key factors, shifting from subjective hypothesis testing and latent-variable path analysis to objective, measured variables insights.

    “In the second stage, we use explainable machine learning techniques to identify key factors, shifting from subjective hypothesis testing and latent-variable path analysis to objective, measured variables insights.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given

    The pattern of results is partially consistent with previous findings on ChatGPT adoption.

    “This pattern is partially consistent with previous findings on ChatGPT adoption.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    The Theory of Planned Behavior posits that an individual's behavioral intention is influenced by three key factors: attitude toward the behavior, subjective norm, and perceived behavioral control.

    “Building on TRA, Ajzen proposed the TPB, which posits that an individual’s behavioral intention is influenced by three key factors: attitude toward the behavior, subjective norm, and perceived behavioral control.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Ajzen, 1991 source

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    In the context of GenAI, perceived usefulness mainly reflects its capacity to assist in mastering theoretical knowledge, building cognitive frameworks, filtering relevant information, facilitating academic learning, supporting foreign-language reading, aiding programming, and fostering interdisciplinary understanding.

    “In the context of GenAI, perceived usefulness mainly reflects its capacity to assist in mastering theoretical knowledge, building cognitive frameworks, filtering relevant information, facilitating academic learning, supporting foreign-language reading, aiding programming, and fostering interdisciplinary understanding (Liu and Zhang, 2025).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Liu and Zhang, 2025

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    The 2SSX framework combines SEM with explainable machine learning techniques, tests theoretical paths, and identifies key factors at a granular level, offering additional insights beyond SEM.

    “Introducing the 2SSX framework, which combines SEM with explainable machine learning techniques. This dual-stage approach not only tests theoretical paths but also identifies key factors at a granular level. The 2SSX framework offers additional insights beyond SEM”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationnone given

    The framework blends constructs from the Technology Acceptance Model, Theory of Planned Behavior, Perceived Risk, and Knowledge Attitude Practice models, emphasizing students' intrinsic motivations.

    “Grounded in an integrated benefit–risk perspective, this framework blends constructs from the Technology Acceptance Model, Theory of Planned Behavior, Perceived Risk, and Knowledge Attitude Practice models, emphasizing students’ intrinsic motivations.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    In the case of GenAI, social risks include concerns over information authenticity, employment displacement, academic misconduct, and technological advancement surpassing regulatory oversight.

    “In the case of GenAI, social risks include concerns over information authenticity, employment displacement, academic misconduct, and technological advancement surpassing regulatory oversight (Hamed et al., 2024; Salari et al., 2025).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Hamed et al., 2024 source

      Not checked in the free run.

    • cites Salari et al., 2025 source

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    This methodological gap is acknowledged, and the XML findings are used to specify salient indicators and practical priorities rather than to draw causal conclusions.

    “This methodological gap is acknowledged, and the XML findings are used to specify salient indicators and practical priorities rather than to draw causal conclusions.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Together, SEM and XML constitute the 2SSX framework, which links mechanism explanation at the latent construct level to factor concretization at the observable indicator level.

    “Together, SEM and XML constitute the 2SSX framework, which links mechanism explanation at the latent construct level to factor concretization at the observable indicator level.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    XML is introduced to concretize the core influencing factors at the indicator level.

    “In Step 2, we introduce XML to concretize the core influencing factors at the indicator level.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The study followed rigorous training protocols in applying machine learning algorithms to ensure optimal performance.

    “To ensure optimal performance, we followed rigorous training protocols in applying machine learning algorithms.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    SEM establishes the theoretically grounded construct-level mechanism, while XML highlights which measured indicators most strongly represent those constructs and are most empirically prominent for adoption-related outcomes.

    “SEM establishes the theoretically grounded construct level mechanism, while XML highlights which measured indicators most strongly represent those constructs and are most empirically prominent for adoption related outcomes.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    SHAP is a game-theoretic approach that decomposes model predictions into additive feature contributions.

    “The second method employed SHAP (SHapley Additive exPlanations), a game-theoretic approach that decomposes model predictions into additive feature contributions (Jia et al., 2024; Qian et al., 2024).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Jia et al., 2024 source

      Not checked in the free run.

    • cites Qian et al., 2024 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    This indicator-level refinement is useful when SEM identifies a significant construct-level effect but does not specify which component items are more decisive.

    “In practical terms, this indicator-level refinement is useful when SEM identifies a significant construct-level effect but does not specify which component items are more decisive.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The model integrates constructs related to both benefits and risks.

    “The model integrates constructs related to both benefits and risks.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductnone given

    By aggregating SHAP values across samples, this approach enables robust importance ranking and reveals potential interaction effects between features.

    “By aggregating SHAP values across samples, this approach enables robust importance ranking and reveals potential interaction effects between features.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The node-splitting analysis method provides a global view of variable importance within the decision tree structure.

    “This method provides a global view of variable importance within the decision tree structure.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationnone given

    The 2SSX framework combines the strengths of SEM with XML, providing a two-step approach that connects construct-level mechanisms with indicator-level factor specification to understanding GenAI adoption.

    “This framework combines the strengths of SEM with XML, providing a two step approach that connects construct level mechanisms with indicator level factor specification to understanding GenAI adoption.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    XML results are interpreted as complementary evidence that sharpens the understanding of "what specifically matters" at the indicator level, rather than as a direct extension of SEM causal claims.

    “Accordingly, XML results are interpreted as complementary evidence that sharpens the understanding of “what specifically matters” at the indicator level, rather than as a direct extension of SEM causal claims.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    These IML methods facilitate a deeper understanding of the model's internal logic, support debugging and optimization processes, and help identify key variables that most strongly influence students' behavioral intentions toward GenAI adoption.

    “These methods facilitate a deeper understanding of the model’s internal logic, support debugging and optimization processes, and help identify key variables that most strongly influence students’ behavioral intentions toward GenAI adoption (Linardatos et al., 2020).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Linardatos et al., 2020 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Using ChatGPT may diminish sustained creative output and lead to content homogenization.

    “Zhou’s natural and longitudinal experimental studies demonstrated that using ChatGPT may diminish sustained creative output and lead to content homogenization (Zhou et al., 2025).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Zhou et al., 2025 source

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    XML does not function as an objective validation of SEM; instead, it serves as an item-level refinement layer that translates construct-level mechanisms into empirically prominent and actionable measured factors.

    “In this sense, XML does not function as an objective validation of SEM. Instead, it serves as an item level refinement layer that translates construct level mechanisms into empirically prominent and actionable measured factors.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Higher perceived social risks related to a product or service may reduce users' willingness to recommend it, particularly under peer influence.

    “Featherman and Pavlou (2003) argued that higher perceived social risks related to a product or service may reduce users’ willingness to recommend it, particularly under peer influence.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Featherman and Pavlou, 2003 source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    SEM is employed as a theory-consistent mechanism model to evaluate how key latent constructs jointly shape GenAI adoption.

    “In Step 1, we employ SEM as a theory consistent mechanism model to evaluate how key latent constructs jointly shape GenAI adoption.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    Globally, SHAP values rank the overall importance of input features; locally, they quantify each feature's marginal contribution to an individual prediction.

    “Globally, SHAP values rank the overall importance of input features; locally, they quantify each feature’s marginal contribution to an individual prediction.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    SEM provides a coherent account of the underlying mechanism implied by the study's theoretical framework.

    “It also provides a coherent account of the underlying mechanism implied by the study’s theoretical framework.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    A pilot test was conducted with 15 participants to ensure content validity and reduce potential misunderstanding of questionnaire items.

    “To ensure content validity and reduce potential misunderstanding of questionnaire items, a pilot test was conducted with 15 participants.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Together, these relationships form an integrated framework that captures both the motivational and inhibitory factors that shape students' adoption behavior.

    “Together, these relationships form a integrated framework that captures both the motivational and inhibitory factors that shape students’ adoption behavior.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Such item-level evidence enables the study to move from a general construct inference to more targeted and actionable recommendations.

    “Such item-level evidence enables the study to move from a general construct inference to more targeted and actionable recommendations”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Typically, individuals acquire new knowledge that shapes their attitudes, which subsequently influence their behavioral actions.

    “Typically, individuals acquire new knowledge that shapes their attitudes, which subsequently influence their behavioral actions (Ritter et al., 2025).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Ritter et al., 2025 source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The interpretable modeling strategies constitute the second stage of the proposed two-step SEM-XML framework.

    “Collectively, these interpretable modeling strategies constitute the second stage of the proposed two-step SEM–XML framework.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    By operating at the construct level, SEM clarifies the conceptual relationships among motivations, attitudes, perceived risks, and adoption intention.

    “By operating at the construct level, SEM clarifies the conceptual relationships among motivations, attitudes, perceived risks, and adoption intention.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    The Harman Single-Factor Test showed that the first principal component explained 34.2% of the total variance, below the 40% threshold, indicating that common method bias is not a concern in this study.

    “The test results showed that the first principal component explained 34.2% of the total variance, which is below the 40% threshold, indicating that common method bias is not a concern in this study.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Study design and conductnone given

    A variable is considered a core influencing factor only when it is identified by two different interpretive methods simultaneously.

    “A variable is considered a core influencing factor only when it is identified by two different interpretive methods simultaneously.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Each construct was measured using several reflective indicators derived from established scales in prior literature, adapted to the context of GenAI use in higher education.

    “Each construct was measured using several reflective indicators derived from established scales in prior literature, adapted to the context of GenAI use in higher education.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    Whereas SEM primarily identifies whether construct-level relations are supported, XML is used to characterize how specific observed indicators account for variation in adoption outcomes in the empirical data.

    “Whereas SEM primarily identifies whether construct-level relations are supported, XML is used to characterize how specific observed indicators account for variation in adoption outcomes in the empirical data.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductnone given

    A positive SHAP value indicates that a feature positively contributes to the likelihood of GenAI usage, whereas a negative value implies an inhibiting effect.

    “A positive SHAP value indicates that a feature positively contributes to the likelihood of GenAI usage, whereas a negative value implies an inhibiting effect.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The machine learning process consists of three key steps: feature input, model construction, and performance evaluation.

    “The machine learning process consists of three key steps: feature input, model construction, and performance evaluation (Qian et al., 2024; Qiu et al., 2025).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Qian et al., 2024 source

      Not checked in the free run.

    • cites Qiu et al., 2025 source

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    The Knowledge-Attitude-Practice model suggests that knowledge serves as the precursor to attitude formation, and attitudes, in turn, drive behavioral practice.

    “The model suggests that knowledge serves as the precursor to attitude formation, and attitudes, in turn, drive behavioral practice (Andrade et al., 2020; Salama et al., 2025).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Andrade et al., 2020 source

      Not checked in the free run.

    • cites Salama et al., 2025 source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The two steps of the 2SSX framework are coupled.

    “The two steps are coupled.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    This systematic integration of path analysis and factors identification strengthens the methodological robustness of the study and yields deeper insights into the behavioral mechanisms underlying GenAI adoption in higher education.

    “This systematic integration of path analysis and factors identification strengthens the methodological robustness of the study and yields deeper insights into the behavioral mechanisms underlying GenAI adoption in higher education.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The XML results are interpreted as indicator-level salience patterns rather than causal confirmation.

    “The XML results are therefore interpreted as indicator level salience patterns rather than causal confirmation.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Tool usability and perceived usefulness play a critical role in students' decisions to adopt GenAI tools.

    “emphasizing the critical role of tool usability and perceived usefulness in students’ decisions to adopt GenAI tools.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The more students understand the tool, the more confident they are in using it, and the more willing they are to adopt new technologies.

    “The more students understand the tool, the more confident they are in using it, and the more willing they are to adopt new technologies.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    The study relies on self-reported data, which may be subject to response bias and might not fully reflect users' actual behaviors and experiences.

    “Second, the study relies on self-reported data, which may be subject to response bias and might not fully reflect users’ actual behaviors and experiences.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    These findings are consistent with those of Zhang and Wang (2025) and Wang et al. (2025).

    “These findings are consistent with those of Zhang and Wang (2025) and Wang et al. (2025).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Zhang and Wang (2025)

      Not checked in the free run.

    • cites Wang et al. (2025) source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    Students are increasingly worried about the authenticity and academic credibility of generated content, which may lead them to use these tools more cautiously or even limit their usage due to fears of academic misconduct.

    “Students are increasingly worried about the authenticity and academic credibility of generated content, which may lead them to use these tools more cautiously or even limit their usage due to fears of academic misconduct.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    All machine learning models were trained and validated on the same dataset to ensure comparability.

    “All models were trained and validated on the same dataset to ensure comparability.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    All constructs achieved CR > 0.70 and AVE > 0.50, confirming adequate convergent validity and internal consistency reliability.

    “All constructs achieved CR > 0.70 and AVE > 0.50, confirming adequate convergent validity and internal consistency reliability.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Perceived ease of use and perceived usefulness both show significant positive effects on students' behavioral intention and on their reported actual use.

    “Perceived ease of use and perceived usefulness both show significant positive effects on students’ behavioral intention and on their reported actual use.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The SEM path coefficients represent causal effects of the predictors on behavioral intention and actual behavior, rather than only associations.
    • Rests on an assumption: The self-reported measures of actual behavior accurately reflect students’ actual usage of GenAI tools.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    These findings are consistent with those of Albayati (2024), Zhang L. et al. (2025), and Zhang X. et al. (2025).

    “These findings are consistent with those of Albayati (2024), Zhang L. et al. (2025), and Zhang X. et al. (2025).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Albayati (2024) source

      Not checked in the free run.

    • cites Zhang L. et al. (2025)

      Not checked in the free run.

    • cites Zhang X. et al. (2025)

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Due to space constraints, the study focused primarily on the analysis and application of the 2SSX framework, limiting the comprehensiveness of the analysis.

    “Additionally, due to space constraints, this study focused primarily on the analysis and application of the 2SSX framework, limiting the comprehensiveness of the analysis.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    All model fit indices meet or exceed recommended standards, except for the slightly lower TLI, indicating overall satisfactory model fit (χ/df = 2.863, RMSEA = 0.043, CFI = 0.923, IFI = 0.954, TLI = 0.895, NFI = 0.933).

    “In Table 7, all model fit indices meet or exceed recommended standards, except for the slightly lower TLI, indicating overall satisfactory model fit ( χ /df = 2.863, RMSEA = 0.043, CFI = 0.923, IFI = 0.954, TLI = 0.895, NFI = 0.933).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Grid search with k-fold cross-validation was employed to identify the best combination of parameters for each model.

    “Grid search with k-fold cross-validation was employed to identify the best combination of parameters for each model.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    The deep tree structure and lack of subsampling may increase the risk of overfitting.

    “However, we acknowledge that the deep tree structure and lack of subsampling may increase the risk of overfitting.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    The usability and intuitiveness of a product will directly influence its popularity and effectiveness.

    “Therefore, the usability and intuitiveness of a product will directly influence its popularity and effectiveness.”
    Quote in the paperNot a substantial claim
75 of 75

  • The paper relies on this without support.

    SEM is typically used to analyze direct, indirect, and moderating relationships between latent variables, but it lacks a deeper examination of measured variables.

    “Moreover, SEM is typically used to analyze direct, indirect, and moderating relationships between latent variables, but it lacks a deeper examination of measured variables.”

  • The paper relies on this without support.

    The Technology Resistance Theory posits that technology adoption decisions depend not only on perceived benefits but also on perceived risks.

    “The Technology Resistance Theory posits that technology adoption decisions depend not only on perceived benefits but also on perceived risks.”

  • The paper relies on this without support.

    Hyperparameter tuning is a critical step in achieving optimal model performance.

    “Hyperparameter tuning is a critical step in achieving optimal model performance.”

  • The paper relies on this without support.

    Subjective norms play an essential role in decision-making, as individuals often experience social pressure when determining whether to engage in a particular behavior.

    “Subjective norms play an essential role in decision-making, as individuals often experience social pressure when determining whether to engage in a particular behavior.”

  • The paper relies on this without support.

    The Knowledge–Attitude–Practice model is one of the most widely used frameworks to explain how individual knowledge and beliefs influence behavioral change, particularly in the context of health behavior research.

    “The Knowledge–Attitude–Practice model is one of the most widely used frameworks to explain how individual knowledge and beliefs influence behavioral change, particularly in the context of health behavior research.”

  • The paper relies on this without support.

    In the context of GenAI, perceived risk can be divided into three categories: Perceived Personal Risk, Perceived Academic Environmental Risk, and Perceived Social Risk.

    “In the context of GenAI, perceived risk can be divided into three categories:”

  • The paper relies on this without support.

    Beyond individual concerns, academic environmental risks also affect adoption intentions.

    “Beyond individual concerns, academic environmental risks also affect adoption intentions.”

  • The paper relies on this without support.

    When individuals begin using new technologies, they often encounter personal concerns that may hinder adoption.

    “When individuals begin using new technologies, they often encounter personal concerns that may hinder adoption.”

  • The paper relies on this without support.

    The SEM path coefficients represent causal effects of the predictors on behavioral intention and actual behavior, rather than only associations.

    “The analysis reveals that Perceived Ease of Use has a significant positive effect on Behavioral Intention (β = 0.100, p < 0.05).”

  • The paper relies on this without support.

    The sample of undergraduate, master’s, and doctoral students from several universities in Southwestern China is representative of university students in general.

    “The questionnaire was distributed to undergraduate, master’s, and doctoral students in several universities across Southwestern China.”

  • The paper relies on this without support.

    The self-reported measures of actual behavior accurately reflect students’ actual usage of GenAI tools.

    “The dependent variable, AB, was represented by the mean of its measured variables.”
11 of 11

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Predictive modelling of GenAI learning outcomes in the literature is dominated by regression and structural equation approaches.

    “Lastly, predictive modelling of GenAI learning outcomes in the literature is dominated by regression and structural equation approaches [1, 52, 67]”

    From Multinational machine learning investigation of whether generative AI interactions augment or substitute cognitive skills and study efficiency in global higher education (2026) · cites this paper as [67]

    The Results and discussion sections on path analysis and ML model construction describe this paper’s use of structural equation modeling and regression algorithms. But they model students’ GenAI adoption and actual usage, not learning outcomes, and the copy does not establish that these approaches dominate the literature. (LLM, conf=0.81)

    Experimental — accuracy not yet measured

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Feature-importance visualization is attributed to the learning-curve figure

    Minor

    The text points to Figure 4 for the top-feature visualization, but Figure 4 is captioned as learning curves; the feature-ranking visualization is captioned Figure 5.

  • CFA includes a perceived-ease-of-use item absent from the questionnaire list

    Minor

    Table 2 lists PE1 through PE4 and then begins the PU items, whereas Table 6 includes PE5. The described removal and renumbering of items does not account for an additional PE item.

  • H1 and H2 are assigned to the opposite constructs in the results

    Minor

    The stated hypotheses assign perceived usefulness to H1 and ease of use to H2. Table 8 reverses those assignments.

  • PSR1 remains despite the stated SMC removal rule

    Minor

    The paper says items with SMC below 0.50 were removed, but Table 6 retains PSR1 with an SMC of 0.493.

  • Machine-learning conclusion changes the modeled outcome

    Minor

    The machine-learning model predicts actual behavior (AB), but the conclusion presents its TA2 finding as significantly driving intention to adopt. Intention was not the stated dependent variable for that analysis.

  • Gender percentages are reversed in the sample description

    Minor

    The description says 51.58% of respondents were male and 48.42% female, but Table 1 reports 48.52% male and 51.48% female.

6 of 6

Cites

Flagged first, then by citations

Research works printed but not linked: 4

  1. Brown, 2015, Confirmatory factor analysis for applied research
  2. Kline, 2023, Principles and practice of structural equation modeling
  3. Singh, 2014, Sampling techniques & determination of sample size in applied statistics research: an overview, Int. J. Econ. Commer. Manag., 2, 1
  4. Tabachnick, 2007, Using multivariate statistics

Cited by

Sorted by citations

Related

Authors
Zeng, D., Xu, X., Zhu, T., Li, Y., & Li, Q.
Affiliations
  • Chengdu University of Information Technology (CN)
  • Jinan University (CN)
  • National Bureau of Statistics of China (CN)
Corresponding author
Li, Yong
ORCID
Li, Qiumin
Venue
Frontiers in Psychology
Publisher
Frontiers Media SA
Published
2026-01-29
Volume
17
Pages
1743722-1743722
ISSN
1664-1078 (ISSN-L)
DOI
10.3389/fpsyg.2026.1743722
PMC
PMC12894297
OpenAlex
W7126038429
Semantic Scholar
be9c6092628128c587ed7b5dc7da344f46932b40
Cited by
7
References
72
Citations by year
2026: 7
FWCI
50.73
Type
journal article
Language
en
Open access
Licence
Locations
  • Frontiers in Psychology (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
Topics
  • Social Sciences › Decision Sciences › Information Systems and Management › Technology Adoption and User Behaviour
  • Physical Sciences › Computer Science › Information Systems › Artificial Intelligence in Education
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
Keywords
generative AI adoption, university students, intrinsic motivation, risk perception, technology acceptance model, theory of planned behavior, knowledge, attitudes and practices, perceived usefulness, academic integrity, ChatGPT, higher education, explainable machine learning

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review