Advancing STEM Education for Sustainability: The Impact of Graphical Knowledge Visualization and User Experience on Continuance Intention in Mixed-Reality Environments

Abstract

Knowledge visualization has gained significant research attention for its potential to facilitate knowledge construction through interactive graphics while minimizing cognitive load during information processing. However, limited research has examined the integration of knowledge visualization within highly interactive mixed-reality environments and its effects on user experiences and science, technology, engineering, and mathematics (STEM) sustainability. Drawing on the cognitive-affective model of immersive learning, this study investigates how learners’ user experiences, elicited by mixed-reality features and usability, influence their sustainable engagement with STEM learning through knowledge-visualization tools framed within the stimulus–organism–response model. A novel mixed-reality learning system was developed, with the user interface designed using concept maps to graphically visualize concept nodes and their interconnected relationships. A total of 136 learners from two high schools in China participated in an experiment on frictional physics using this novel system. Using structural equation modeling, the collected data were analyzed with partial least squares. The findings demonstrate that mixed-reality features of knowledge visualization (featured by 3D graphics, interface design, and operational functions), as well as usability (featured by the perceived usefulness of the concept map, perceived ease of use, and perceived usefulness of the system), have positive significant impacts on user experience (represented by satisfaction, perceived enjoyment, and attitude). Subsequently, positive user experiences have positive significant impacts on learners’ sustained intention to engage with STEM education. Further mediating analysis provides empirical evidence that positive user experiences, acting as a psychological enabler, mediate the relationship between system design and behavioral intention. The research model explains 65.2% of the variance for system usability, 53.4% for satisfaction, 51.5% for perceived enjoyment, 54.9% for attitude, and 63.2% for continuance intention. By fostering positive user experiences in STEM learning, this study offers valuable insights for educators and practitioners seeking to implement effective interactive knowledge visualizations to support sustainable STEM education and immersive learning.

1. Introduction

In the context of an increasingly challenging global economic situation and fierce social competition, particularly exacerbated by the global spread of COVID-19, there is an urgent need for the younger generation to adapt to new economic, political, and social realities and to be adequately prepared for their future careers. To enhance international influence and remain competitive in the global knowledge economy, countries such as China, Russia, India, the United States, and Australia have placed greater emphasis on science, technology, engineering, and mathematics (STEM) education as a critical driver of future economic prosperity [1,2,3]. Beyond its economic and political significance, STEM education plays a key role in developing student competencies, offering metacognitive learning strategies that help students reflect on their cognitive processes and apply strategic methods to solve real-world challenges. The integration of mobile devices, such as smartphones, laptops, and tablets, plays a pivotal role in enhancing STEM education by facilitating intelligent scaffolding mechanisms that support the efficient organization and assimilation of complex learning information [4]. Previous research has demonstrated that computer-supported learning environments offer significant advantages over traditional settings, including improvements in students’ theoretical knowledge, applied abilities, and scientific inquiry competencies [5,6,7,8]. By promoting flexible, technology-enhanced learning environments, the digital learning systems contribute to the sustainability of education, ensuring equitable access to knowledge and fostering continuous engagement with STEM disciplines in an increasingly digitalized world [9]. Additionally, sustained continuance intention in STEM subjects is crucial for reinforcing knowledge retention, fostering critical thinking skills, and enhancing students’ perseverance and adaptability in tackling complex problems [10].

Among emerging technologies, mixed reality (MR) has shown great promise in enhancing STEM education [11]. MR as an integrated learning modality has shown considerable promise in cultivating lifelong STEM proficiency [12]. It is widely agreed among researchers that when pedagogical resources are seamlessly integrated with real and virtual objects, MR—through its ability to construct immersive learning environments (MRLEs)—provides rich teaching methods, content, and opportunities to enhance learners’ problem-solving skills and conceptual understanding [13,14,15]. In STEM education, MR technology enables the creation of interactive learning environments, allowing students greater autonomy in their learning activities, which can foster increased independence. However, in the context of complex STEM education systems and vast amounts of abstract information, learners often experience disorientation and cognitive overload [16,17,18]. In MR-based learning, learners are typically required to navigate virtual spaces to acquire essential information, often within a decentralized structure that complicates the formation of meaningful knowledge structures, particularly for novices. This overwhelming cognitive load, including tasks such as locating learning cues, processing new information, and integrating new knowledge with existing concepts, can place significant strain on learners’ working memory in a short time frame, leading to what has been referred to as being ‘Lost in Hyperspace’ [19,20,21]. Given the challenges learners encounter in navigating complex and disordered systems—often stemming from disorientation and cognitive overload—there is a pressing need for research dedicated to developing effective and sustainable scaffolding tools that facilitate cognitive efficiency and knowledge navigation in MR learning environments. One promising approach to addressing this challenge is knowledge visualization (KV), a technique that transforms abstract concepts and their interrelationships into interactive visual representations. Prior research has demonstrated that well-structured visualizations can reduce extraneous cognitive load and support meaningful learning by clarifying conceptual relationships [22,23]. Importantly, effective visual design also contributes to a more positive user experience by enhancing perceived clarity, enjoyment, and learning satisfaction [24,25], which are particularly important for sustaining motivation and intention in STEM learning contexts.

Given the limitations of working memory, learners need efficient scaffolding strategies to facilitate the learning process. Among various approaches, cognitive and metacognitive learning strategies have been identified as key to supporting students in completing their learning tasks effectively [26,27,28]. The cognitive-affective model of immersive learning (CAMIL) [29] emphasizes the role of emotional and cognitive processes in immersive learning, highlighting how positive affective experiences can enhance involvement and learning outcomes in immersive environments. Additionally, the Control-Value Theory of Achievement Emotions (CVTAE) [30] explains how emotions, such as enjoyment, frustration, and motivation, affect learners’ persistence in educational tasks, particularly in complex subjects like STEM. These affective-related theories are crucial in understanding how learners’ emotional responses to MR learning systems influence their continued intention and willingness to persist in STEM education. Furthermore, MR-equipped learning systems can attract students by offering enjoyable and interactive experiences, sometimes blurring the line between learning and play. Researchers in user experience (UX) have increasingly emphasized the hedonic dimensions of system use, including enjoyment, affective response, and positive attitude [25,31]. Hedonic user experience pertains to the emotionally gratifying aspects of interacting with a system, which evokes pleasure, enjoyment, and intrinsic motivation. These affective responses are particularly critical in STEM learning contexts, where positive affective states can enhance learners’ sustained interest, deepen cognitive involvement, and ultimately support educational sustainability.

While cognitive and metacognitive strategies suggest that STEM learning is achievable when learners can understand, control, and manage the learning process [18], existing research lacks sufficient evidence to demonstrate the pedagogical value of MR in improving students’ UX in STEM fields. Consequently, further research is urgently needed to explore effective methods for building MR-based learning environments and to examine the specific pedagogical benefits that MR technology can offer for sustainable STEM education. To achieve the research objective, this study explores a novel MR learning system that utilizes KV as a scaffolding tool. The knowledge structure is visually represented by extracting key concept nodes and their logical relationships from the syllabus. Following the WYSIWYG paradigm [32,33], the KV design in this study is concise and serves as a cognitive roadmap to guide learners through knowledge exploration in a complex learning system, thereby supporting conceptual knowledge construction. Separate MR learning scenarios are then developed based on the corresponding concept nodes, offering a resource-rich MR learning environment where learners can engage in hands-on experiments with both real and digital content in a learning-by-doing approach [34,35]. Additionally, a knowledge-check-based question-answering module is integrated into the MR system’s storyline to encourage the development of learners’ metacognitive thinking skills [36], specifically their ability to summarize and self-assess the conceptual knowledge they have learned. This paper develops a conceptual model based on the stimulus-organism-response (S-O-R) framework to explore how the features of a KV-based MR system influence learners’ hedonic user experience and, ultimately, impact their STEM continuance intentions.

2. Theoretical Background and Hypothesis Development

2.1. The Stimulus–Organism–Response Framework in Immersive Learning Technologies

The S-O-R framework, originally derived from environmental psychology by Mehrabian [37], has gained widespread application in analyzing user behavior in response to external environmental stimuli across diverse domains. Recent research has demonstrated the utility of the S-O-R framework in educational contexts, such as online learning [38,39], Massive Open Online Courses (MOOCs) [40,41], higher education [42,43], and gamification [44,45]. According to the model, when an individual is exposed to environmental stimuli (S), this triggers an internal psychological state (organism, O), which subsequently influences both their cognitive and affective processes. These processes, in turn, lead to a behavioral response (R) to the stimuli [44].

In this framework, Stimulus (S) refers to the external environmental factors that impact an individual’s cognitive or emotional state. These factors can be varied, ranging from physical and social contexts to digital stimuli, all of which influence the way users engage with systems or experiences. It is essential to select these stimuli carefully, taking into account the specific characteristics of the environment under investigation, as they have a profound effect on user engagement and emotional responses [46].

The organism (O) represents the individual’s internal state, encompassing both cognitive and affective dimensions [47]. The cognitive component pertains to the mental processes involved in acquiring, processing, retaining, and retrieving information, which can be viewed from an information-processing perspective [48]. The affective component refers to emotional responses such as enjoyment, satisfaction, or frustration, which are crucial to understanding user engagement in educational settings [47].

Finally, the response (R) refers to the behavioral outcome resulting from the interplay between the stimulus and the organism’s internal state. This response represents the external manifestation of cognitive and emotional states, such as the decision to engage further with the content, the amount of time spent on a task, or the likelihood of adopting certain behaviors, like sharing or recommending a learning experience.

In the following sections, careful consideration of the specific stimuli, cognitive and affective processes, and observable responses is presented to offer valuable insights for designing more effective and engaging MR learning environments.

2.2. Graphical Visualization of Knowledge Structure in MRLEs

KV has garnered substantial scholarly attention as an effective methodology within the realms of knowmetrics and mental semantic network theory, where it plays a crucial role in enhancing cognitive comprehension and facilitating long-term memory retention among learners [23]. By leveraging visual representations of knowledge, KV facilitates the organization and retrieval of information, which is crucial for externalizing and eliciting abstract cognitive structures. Moreover, KV enhances learners’ abilities to synthesize disparate pieces of information into a cohesive structure within complex educational contexts, such as blended, virtual, and MOOC settings [49,50]. Concept maps have been widely recognized as essential KV techniques for graphically representing the relationships between concept nodes within a hierarchical layout, supporting learners in constructing knowledge and fostering the development of critical thinking skills [51,52]. Although advancements in educational technologies have significantly promoted KV methodologies, such as concept maps, conceptual models remain scarce for uncovering how to effectively integrate pedagogical functions with visual representations and whether knowledge construction has been sufficiently fostered [53]. In this study, a mobile MR system integrated with KV functions, implemented through interactive concept maps, was developed. The pedagogical functions of the system, characterized by MR learning system features and usability, were designated as the stimuli constructs within the S-O-R research model.

2.2.1. KV-Based MR Learning System Features

The mobile MR learning system leverages the integration of computer vision and device-embedded sensors to superimpose three-dimensional (3D) graphics—such as objects, images, and animations—onto real-world environments in real-time [54]. According to Azuma [55], AR technologies are defined by three fundamental criteria: (1) the seamless combination of real and virtual environments, (2) the facilitation of real-time interactions, and (3) precise registration of virtual content within a 3D spatial framework. These principles underscore the importance of integrating 3D graphics with maker-based mobile MR technologies to enhance users’ embodied experience by bridging the digital and physical worlds. Under the cognitive theory of multimedia learning [56], the design elements of MR interventions can significantly enhance learners’ cognitive processing by promoting the integration of visual and auditory information. These design elements, such as spatial annotation, vision-haptic visualization, and 3D multimodal representations [57], are strategically designed to minimize cognitive overload while maximizing the efficient use of working memory. Consequently, 3D graphics serve as critical design elements that encapsulate the core features of MR learning systems.

The interface design of MR systems has been widely recognized by researchers as a critical factor in enhancing system usability and sustaining high-quality learning experiences for learners [15,58,59]. Effective interface design not only facilitates intuitive interactions but also minimizes cognitive load, enabling learners to focus on the educational content. Specifically, the concept map, recognized as an efficient knowledge visualization (KV) approach, has been employed as a user interface (UI) in mixed-reality (MR) learning applications [60,61]. This approach demonstrated significant user-friendliness during experimental studies with learners, enabling intuitive navigation of complex information structures. These findings suggest that a well-designed interface facilitates participants in effectively gathering information and constructing knowledge, thereby fostering a deeper level of engagement and motivation among learners [62].

Operational functions have undergone significant evolution alongside the advancements in MR technology. With continuous improvements in hardware capabilities, such as enhanced processing power, more precise sensors, and higher-resolution displays, MR systems now support more sophisticated interactions and seamless integration of virtual and physical elements. Furthermore, advancements in software algorithms, including real-time spatial mapping, gesture recognition, and artificial intelligence, have expanded the scope of MR applications across diverse fields such as education, healthcare, and engineering [63]. From the instructional-operational perspective, reliable operation functions minimize disruptions during learning activities and increase individuals’ willingness to actively engage with and support sustainable STEM learning paradigms in MRLEs [54].

Therefore, this study developed a second-order construct, referred to as MR features, by integrating three key constructs associated with the features of KV in MR learning systems: 3D graphics, interface design, and operational functions [62].

2.2.2. KV-Based MR Learning System Usability

The continued adoption of novel technology, such as MR, is closely linked to an individual’s personal experience throughout the learning trajectory, particularly to whether the system’s functionality is perceived as useful [10]. System usability, grounded in cognitive ergonomics, refers to the degree to which a system enables effectiveness, efficiency, and user satisfaction in achieving goals within human–computer interaction (HCI) contexts [64,65]. Usability in MRLEs is an emergent property characterized by ease of use and user-friendliness, which manifests through interactions among users, tools, tasks, and real-virtual environments.

Devis [66] proposed the technology acceptance model (TAM) to explore the influence of perceived usefulness (PU) and perceived ease of use (PEU) on users’ learning perception (i.e., attitude, enjoyment) and behavioral intentions. Specifically, PEU refers to the degree to which learners believe that using the technology will be free of effort, while PU reflects their belief that the technology will enhance their learning experience or task performance. Grounded in the psychology-based theory of reasoned action, TAM provides a robust theoretical foundation for understanding how PU, PEU, and other constructs contribute to the behavioral intention and attitude of students and educators toward the continued use of immersive technologies in STEM learning contexts, as highlighted by Nilashi and Abumalloh [67]. For instance, TAM has been effectively applied in practical MR applications to examine learners’ PEU, PU, and attitude toward use within the cultural heritage domain [68]. Additionally, Sepasgozar [69] states that PU and PEU emerge as pivotal predictors of participants’ satisfaction within immersive virtual trips in the context of engineering training.

Regarding concept map tools, Whitelock-Wainwright et al. [70] consider the perceived usefulness of concept mapping tools to serve as a valid proxy for metacognition in the process of knowledge construction. When learners engage deeply with these pedagogical tools, their understanding of the effectiveness of concept maps is enhanced, which in turn informs and facilitates their future choices regarding tool usage. Prasetya et al. [71,72] validated the suitability of TAM in assessing the learners’ perceptions of using concept maps for building knowledge structure. However, despite the attention given to usability by scholars in the immersive research field and concept maps, the usability of concept maps as a KV approach in MRLEs has not been adequately investigated.

Therefore, this study created a second-order construct, usability, combining three constructs relating to the usability of the MR learning system and concept maps, with PU and PEU focusing on usability at the system level and perceived usefulness of concept map (PUCM) specifying the usability of KV approach in MRLEs.

2.3. User Experience

User experience encompasses a wide range of factors associated with product or system use, emphasizing subjective perceptions and interactions [73]. Beyond classical usability’s focus on instrumental factors like efficiency and effectiveness, UX highlights how these attributes influence emotional responses, such as motivation, satisfaction, and presence. Researchers highlight the significance of UX measurement in the field of HCI, which captures the uniqueness and variation of user experiences with technology and offers guidelines for enhancing learning outcomes and promoting sustained end-user system usage [74,75]. In immersive learning contexts, UX is shaped by the interplay of several factors, including contextual components (e.g., users, devices, and interaction modes), technological affordances (such as the ability to create a seamless blend of virtual and physical realities), and the inherent characteristics of immersive systems (e.g., potential side effects like motion sickness or sensory fatigue) [76]. To evaluate the UX of learners interacting with the KV-based MR system, this study incorporates three constructs that capture key aspects of UX: satisfaction, perceived enjoyment, and attitude.

Satisfaction (SF) is conceptualized as a cognitive construct that reflects the alignment or misalignment of learners’ expectations with their actual experiences in educational systems or interventions. Grounded in interdependence theory, SF is instrumental in shaping the reciprocal relationship between learners and educational platforms [77]. Preliminary research indicates that satisfaction significantly contributes to usage continuance intention and learning stickiness in location-based augmented reality environments [78]. Similarly, Lim et al. [79] demonstrate that the PEU, innovativeness, and PU of immersive systems play a pivotal role in enhancing SF and user inclination to adopt these technologies.

Perceived enjoyment (PE) refers to the degree to which learners find the hedonic aspects of the system intrinsically rewarding to use [80]. PE is widely incorporated as a crucial mediator reflecting intrinsic motivation and overall pleasure with innovative systems, independent of performance-related outcomes, thereby significantly shaping future technology adoption. In immersive learning contexts, users who derive enjoyment from their learning journey in virtual reality environments are more likely to form an affirmative stance on the technology, as suggested by research conducted by Jo and Park [81]. Holdack et al. [82] validate the mediating role of PE towards PU in AR applications by facilitating cognitive processing, which in turn enhances attention and learning outcomes and fosters innovation. This mediating role underscores the interconnectedness of affective and cognitive dimensions in shaping users’ perceptions of technology and influencing their behavioral intentions.

Attitude (AT) refers to the learners’ positive or negative evaluation of the system, which influences their beliefs to perform the target behavior [83]. Attitudinal learning assessments have been increasingly employed in recent pedagogical research to evaluate the impact of technology-enhanced learning systems on students’ attitudes across various learning contexts, particularly in immersive environments [84]. Adopting a holistic perspective on attitudes toward novel technologies provides deeper insights into the relationship between users’ attitudinal shifts and the instructional design of immersive learning. Evidence suggests that the affordances of immersive technology, coupled with its intrinsic interactivity, effectively shape user attitudes, subsequently motivating sustained engagement [81]. Furthermore, the experiential and situated learning strategies employed in immersive learning environments cultivate positive attitudes, enhance critical thinking during reflective practice, and strengthen professional learning processes by enabling pre-service teachers to reinterpret their experiences and gain deeper insights into their actions [85].

Several studies underscore the significance of UX in immersive learning and its influence on STEM continuance intentions. With this backdrop, the present study endeavors to delve into the construction of UX as the organism component in the S-O-R framework, which encompasses three main constructs: PE, SF, and AT. These constructs serve as mediators, elucidating the role of UX in linking external stimuli to learners’ behavioral responses in KV-based MRLEs.

2.4. STEM Continuance Intention in Immersive Learning Environments

STEM continuance intention (CI) describes students’ willingness to engage in STEM-related activities, adopt STEM tools, or pursue STEM disciplines beyond initial exposure [10]. This construct is pivotal in educational research, as it captures the ongoing interest and involvement necessary for long-term success in STEM fields. Importantly, continuance intentions bridge the gap between short-term engagement and enduring commitment, influencing not only individual learning outcomes but also the cultivation of a future-ready STEM workforce [86]. In the context of immersive learning, numerous studies examine the determinants of CI, emphasizing the alignment between the affordances of the learning system and the goals of the learners. Choi et al. [87] furnished evidence asserting that the relationship between behavioral intention toward the metaverse and AT is moderated by psychological distance. Faqih [88] evaluated the intention to adopt mobile AR games based on nine factors and confirmed that perceived innovativeness significantly influences users’ continued intention to engage with mobile AR technology. Dino et al. [65] employ the Unified Theory of Acceptance and Use of Technology (UTAUT) to investigate the continuance intention of older adults engaging in an MR exercise program. Their empirical findings indicate that the continuance intention for MR technology is shaped by key determinants, including performance expectancy, effort expectancy, social influence, facilitating conditions, and the usability of MR systems.

The present study developed an MR learning system focusing on frictional concepts in physics, which serves as a representative case of STEM education. The construct of STEM CI is conceptualized as the behavioral response within the S-O-R framework, evaluating the extent to which learners exhibit an intention to continue adopting the KV-based MR system as a tool for constructing new knowledge.

2.5. Research Model and Hypotheses

While prior research has extensively applied UX within the S-O-R framework to investigate how technological affordances and UI design shape behavioral intention [89,90], scholarly attention remains disproportionately limited in addressing UX within KV-based immersive learning ecosystems. In this study, KV-integrated MR system characteristics represented by 3D graphics, interface design, operational functions, and system usability, especially regarding the usability of concept maps, are constructed to explain the intertwined relationship influencing UX. The learning habits associated with STEM disciplines are reinforced through individuals’ experiences in adopting immersive learning systems [91]. More importantly, MR is powerful in both formal and informal environments due to its ability to seamlessly blend virtual and physical elements, providing a safe, interactive environment for trial and error through hands-on experimentation with digital objects in real-world contexts. Learners are more likely to demonstrate responsible behavior in completing learning tasks and are less inclined to drop out when placed in environments that offer the flexibility and convenience necessary for facilitating the intuitive understanding of complex concepts [92]. Hence, learners’ perceived UX in immersive environments can be cultivated when they are exposed to contexts that integrate interactive elements, provide intuitive UI, and support the construction of knowledge about complex concepts in a dynamic and engaging manner, thereby fostering affirmative experiences such as satisfaction, enjoyment, and positive attitude. With the provision of high-quality UX, learners are more likely to form positive behavioral intentions toward engaging with immersive learning systems. Drawing from existing literature on the relationship between UX and behavioral intention in immersive environments, a conceptual model for this study (see Figure 1) is presented, along with the proposed hypothesized relationships.

Hypothesis 1 (H1).

KV-based MR features have positive direct relationships with (a) SF, (b) PE, (c) AT, (d) Usability, and (e) CI.

Hypothesis 2 (H2).

KV-based MR usability has a positive direct relationship with (a) SF, (b) PE, (c) AT, and (d) CI.

Hypothesis 3 (H3).

There is a positive direct relationship between (a) SF, (b) PE, (c) AT, and CI.

Hypothesis 4 (H4).

(a) SF, (b) PE, and (c) AT mediate the relationship between MR features and CI.

Hypothesis 5 (H5).

(a) SF, (b) PE, and (c) AT mediate the relationship between MR usability and CI.

3. Methods

3.1. Design of KV-Based MR Learning System

To evaluate the proposed research model, a KV-based MR learning system was developed in this study. The system architecture diagram of the MR prototype is illustrated in Figure 2. The key frictional concepts from the syllabus were identified as concept nodes, and the corresponding relationships between nodes were established by two experienced physics teachers with over ten years of teaching experience. Subsequently, a radical multi-hierarchical concept map [93] was constructed as the MR user interface, with concept nodes represented as 3D spheres and relationships depicted as tagged links with arrows, as shown in Figure 3. The concept nodes situated proximate to the center of the map exhibit a higher degree of abstraction, whereas those positioned toward the periphery demonstrate greater specificity. A detailed walkthrough of the MR interaction and interface is provided using one example—Concept Node 6, which consists of four sub-scenarios.

• Concept node highlighting. The specific nodes situated at the outermost level of the multi-hierarchical concept map are initially locked, necessitating learners unlock them by engaging in hands-on experiments within the MR learning space that seamlessly integrates virtual and physical elements, as illustrated in Figure 3.
• MR Scenario Entering. Specifically, each locked concept node in the concept map corresponds to a dedicated main MR learning scenario. By clicking on a concept node, learners enter the associated main MR environment, which consists of 2 to 4 sub-scenarios. These MR sub-scenarios are designed based on the principles of the controlled variable method. For instance, by altering a specific variable, such as the roughness of a tabletop, students manipulate virtual elements to change the motion of an object, observe the resulting phenomena, and thereby explore the physical concept, as illustrated in Figure 4.
• MR interaction and knowledge check. In each sub-scenario, as illustrated in Figure 5:
  • Learners use a virtual joystick (bottom left of the screen) to apply virtual forces to an object on a mixed-reality surface.
  • Visualized force vectors, representing all individual forces acting on the object (e.g., friction, gravity, normal force, applied force), are superimposed on the virtual objects to enable intuitive observation of dynamic interactions. The system status panel, located at the top left of the screen, provides real-time feedback on object displacement and velocity.
  • Learners observe these dynamics and complete a knowledge-check test (top right). If the answer is incorrect, they must repeatedly experiment until the correct conceptual understanding is formed and the test is passed.
• Concept node unlocking. Once all knowledge-check tests associated with a concept node are successfully completed, the node is unlocked on the concept map, visually confirming the learner’s mastery, as illustrated in Figure 6.

The MR experiment process demonstrates how this system combines immersive visualization, active experimentation, real-time data feedback, and user interaction to support a conceptual understanding of physics. The KV-based MR system implements a gamified unlocking architecture designed to foster a positive attitude and promote their sustained willingness to engage in STEM learning. This design leverages gamification strategies to enhance enjoyment, wherein learners advance through progressive levels by demonstrating mastery of fundamental concepts, thereby cultivating intrinsic motivation and a heightened sense of satisfaction. Additionally, STEM design principles were seamlessly integrated into the development of this innovative MR learning system.

• Science. The MR system is centered on the study of frictional physics, aligning directly with the science dimension of STEM education. Learners are required to understand fundamental physical principles related to concept nodes through the immersive presence and authentic visualization of abstract concepts within the MR environment, fostering a spirit of scientific inquiry.
• Technology. The MR technology provides learners with an innovative and immersive learning experience in high school subjects. By leveraging the capabilities of computer vision, computer graphics, and multi-sensor fusion technologies, the system ensures precise registration and dynamic tracking, allowing for real-time spatial awareness and seamless interaction with virtual elements.
• Engineering. In order to unlock the outermost concept node, learners must complete knowledge quizzes that relate the learned concepts to real-life applications of friction, especially the impact of friction in engineering applications. The knowledge-check-based design facilitates active learning by reinforcing the connection between theoretical concepts and practical applications, ensuring that learners can integrate and apply their understanding of friction in real-world engineering contexts [94].
• Mathematics. The motion and force states of the virtual objects in this MR system are dynamically calculated using mathematical formulas from physics. This real-time visualization, grounded in mathematical principles, helps learners intuitively grasp friction and force concepts by explicitly visualizing the forces acting on objects, providing a clearer and more interactive learning experience, as shown in Figure 5.

The system was developed using marker-based tracking, which is supported by ARFoundation 4.1.0 in Unity 3D 2019.4.29. It was deployed on both Android and iPad platforms through ARCore and ARKit SDKs, respectively, to ensure stable and seamless cross-platform performance. By leveraging these industry-standard frameworks, we aimed to enhance real-time tracking accuracy, optimize rendering efficiency, and maintain a consistent user experience across different devices.

3.2. Population and Experiment Procedure

In this study, a purposive sampling approach was employed to select participants most likely to provide valuable insights into the research question. Purposive sampling, a non-probability technique, involves selecting participants based on specific characteristics or criteria, enabling the researcher to focus on individuals who are most capable of offering relevant and meaningful data related to the research objectives [95]. By targeting high school students with a specific background in frictional physics and MR-based learning systems, the purposive sampling method ensured the selection of individuals who could offer pertinent and contextually rich data. The target population consisted of 136 participants (70 males, 66 females) from two high schools in China, with an average age of 17.2 years (SD = 1.2), ranging from 15 to 19 years.

Participants were invited to engage in experiments using the proposed KV-based MR learning system. All participants received a consent form along with a Participant Information Sheet (PIS), ensuring that they had a comprehensive understanding of the study before providing their consent to participate. As part of the experimental procedure, participants were familiarized with the system and its functionalities. The training process consisted of providing clear, step-by-step instructions to ensure that learners comprehended how to interact with the MR system and its concept map interface. The experiment was conducted in students’ regular classroom environment to simulate a realistic learning context. The procedure was as follows: first, the experiment instructor gave a brief introduction to how to operate the mobile MR device. Then, students accessed the KV-based MR learning system by scanning a marker, which allowed them to enter the MR learning space and begin the experiment. With the assistance of the concept map embedded in the system, students were able to sequentially unlock conceptual nodes and complete the learning tasks in order. This structured guidance minimized the need for external intervention during the experiment. Unless technical issues occurred, the instructor did not interfere with the student learning process. The entire experimental session lasted approximately 100 min.

To ensure a broader sample, participants were organized into groups of one to four individuals, with each group using a single mobile device during the experiment. Learners took turns manipulating the device to ensure equal opportunities for all to interact with the system. Meanwhile, group members could collaborate by discussing concepts and providing timely support to help each other complete learning tasks and unlock the concept nodes. The concept map includes 14 locked concept nodes at the outermost layer, each of which requires learners to complete 2 to 4 MR sub-scenarios designed around key physics concepts using the controlled variable method. This results in dozens of sub-scenarios focused on the topic of friction, allowing everyone sufficient time to experience the system while still maintaining the collaborative element in larger groups. Upon completing the concept map by unlocking all concept nodes, participants were asked to fill out the questionnaires, which were distributed through the web-based survey platform SoJump (https://www.wjx.cn/). They did so either on their personal mobile devices or on computers provided by the experiment staff. The use of an online questionnaire platform ensured standardized data collection and ease of access for all participants.

3.3. Instruments and Measures

The research model consists of 12 constructs, with all measurement items for these constructs derived from existing literature relevant to the research theme. Given that the participants in this study were from two high schools in China, a forward-backward translation method was employed to convert the primary English questionnaire items into Chinese while maintaining the original meaning and ensuring the cultural appropriateness of the translated items [96]. As shown in Figure 1, the measurement items were drawn from the following scales. For the KV-based MR features, three items related to 3D graphics, three items on interface design, and three items on operational functions were adapted from Hao and Lee [62]. For the usability of the KV-based MR system, three items related to the perceived usefulness of the concept map were adapted from Whitelock-Wainwright et al. [70], and four items on PU and four items on PEU were adapted from Venkatesh and Bala [97]. Regarding the participants’ UX during experiments with the KV-based MR learning system, three items on SF were adapted from Ibiliet al. [98], three items on PE were adapted from McLean and Wilson [99], and three items on AT were adapted from Wu et al. [10]. The STEM continuance intention as the response to the usage of the KV-based MR learning system was evaluated using three items adapted from Huang et al. [100] and Wu et al. [10]. The complete instruments for the constructs in the research model are provided in the Supplementary Materials.

The measurement items collected in this study were measured using a 7-point Likert scale, ranging from 1 (strongly disagree) to 7 (strongly agree), to capture the varying degrees of participants’ agreement with the statements. The analysis was performed using Partial Least Squares Structural Equation Modeling (PLS-SEM), a statistical method designed to examine complex relationships among latent variables. The data analysis was conducted with SmartPLS 4.0 software, which is specifically tailored for PLS-SEM applications. PLS-SEM is particularly advantageous for small sample sizes and non-normally distributed data. It allows for the estimation of both direct and indirect effects, making it suitable for exploratory research and complex models with multiple constructs. The process involves two main steps: first, evaluating the measurement model to assess the reliability and validity of the constructs; second, testing the structural model to examine the causal relationships between these constructs [101].

4. Results

The results of the PLE-SEM analysis were obtained through the following steps. First, we examined common method bias (CMB) to ensure the validity of the collected data. Subsequently, we assessed the measurement model, with the results presented in

Table 1. Measurement model assessment on reliability, validity, and collinearity.

Scale and Items	Factor Loadings	Cronbach’s Alpha	CR	AVE	Outer VIF
MR features		0.887	0.931	0.817
3D graphics	0.856	0.814	0.890	0.729	1.953
TDG1	0.867				1.902
TDG2	0.859				1.814
TDG3	0.835				1.696
Interface design	0.936	0.834	0.900	0.751	2.959
ID1	0.877				2.000
ID2	0.838				1.778
ID3	0.884				2.112
Operational functions	0.916	0.778	0.872	0.695	2.520
OF1	0.842				1.898
OF2	0.757				1.384
OF3	0.897				2.163
MR usablity		0.860	0.914	0.781
PUCM	0.879	0.864	0.917	0.786	2.012
PUCM1	0.861				1.980
PUCM2	0.893				2.370
PUCM3	0.906				2.479
PEU	0.875	0.843	0.895	0.680	2.171
PEU1	0.790				1.775
PEU2	0.849				2.115
PEU3	0.845				2.013
PEU4	0.815				1.877
PU	0.905	0.882	0.919	0.740	2.490
PU1	0.846				2.257
PU2	0.907				3.088
PU3	0.882				2.618
PU4	0.802				1.847
SF		0.736	0.848	0.652
SF1	0.840				1.487
SF2	0.699				1.381
SF3	0.872				1.781
PE		0.859	0.914	0.780
PE1	0.870				2.123
PE2	0.880				2.039
PE3	0.899				2.462
AT		0.878	0.925	0.804
AT1	0.900				2.374
AT2	0.890				2.316
AT3	0.901				2.595
CI		0.819	0.892	0.734
CI1	0.804				1.586
CI2	0.881				2.083
CI3	0.884				2.049

Note: PUCM = perceived usefulness of concept map; PEU = perceived ease of use; PU = perceived usefulness; SF = satisfaction; PE = perceived enjoyment; AT = attitude; CI = continuance intention; CR = construct reliability; AVE = average variance extracted; VIF = variance inflation factor. Second-order constructs are represented in bold, while first-order constructs are presented in italics.

and

Table 2. Discriminant validity assessment using HTMT.

Construct	TDG	ID	OF	PUCM	PEU	PU	SF	PE	AT	CI
MR features
TDG
ID	0.831
OF	0.807	0.844
MR Usability
PUCM	0.741	0.763	0.792
PEU	0.728	0.862	0.850	0.734
PU	0.757	0.718	0.768	0.800	0.825
SF	0.758	0.779	0.828	0.779	0.713	0.767
PE	0.746	0.707	0.710	0.745	0.652	0.712	0.799
AT	0.748	0.669	0.682	0.764	0.695	0.768	0.823	0.769
CI	0.665	0.651	0.670	0.693	0.594	0.700	0.867	0.827	0.833

Note: TDG = 3D graphics; ID = interface design; OF = operational functions; PUCM = perceived usefulness of concept map; PEU = perceived ease of use; PU = perceived usefulness; SF = satisfaction; PE = perceived enjoyment; AT = attitude; CI = continuance intention. Second-order constructs are represented in bold, while first-order constructs are presented in italics.

. Following this, we evaluated the structural model and conducted a mediation analysis, the outcomes of which are reported in

Table 3. Mediation analysis.

Total Effects			Direct Effects			Hypothesis	Indirect Effect						Support
β	t-Value	p	β	t-Value	p		β	SE	t-Value	p	LLCI	ULCI
0.614 ***	8.997	0.000	0.007	0.062	0.951	H4a: MRfeatures→SF→CI	0.115 *	0.054	2.117	0.034	0.023	0.230	Yes
0.614 ***	8.997	0.000	0.007	0.062	0.951	H4b: MRfeatures→PE→CI	0.097	0.058	1.670	0.095	0.013	0.232	No
0.614 ***	8.997	0.000	0.007	0.062	0.951	H4c: MRfeatures→AT→CI	0.053	0.043	1.219	0.223	−0.011	0.158	No
0.425 **	3.131	0.002	0.019	0.159	0.874	H5a: MRusability→SF→CI	0.109 *	0.054	2.033	0.042	0.027	0.236	Yes
0.425 **	3.131	0.002	0.019	0.159	0.874	H5b: MRusability→PE→CI	0.126 *	0.064	1.967	0.049	0.030	0.276	Yes
0.425 **	3.131	0.002	0.019	0.159	0.874	H5c: MRusability→AT→CI	0.170 *	0.071	2.395	0.017	0.061	0.337	Yes

Note: * p < 0.05, ** p < 0.01, *** p < 0.001, 10,000 sample bootstrap. Standardized regression coefficients ($\beta$) are reported, with significance levels indicated by asterisks. SF = satisfaction; PE = perceived enjoyment; AT = attitude; CI = continuance intention; LLCI = lower limit of the confidence interval; ULCI = upper limit of the confidence interval.

and

Table 4. Summary of PLS-SEM path analysis.

Path	Path Coefficients	t-Value	p-Values	Decision	Inner VIF	${\mathit{R}}^2$
H1a: MR features→SF	0.393 **	3.176	0.002	Supported	2.873	0.534
H1b: MR features→PE	0.375 **	3.293	0.001	Supported	2.873	0.515
H1c: MR features→AT	0.182	1.524	0.128	Not supported	2.873	0.549
H1d: MR features→MR usability	0.807 ***	24.149	0.000	Supported	1.000	0.652
H1e: MR features→CI	0.007	0.062	0.951	Not supported	3.315	0.632
H2a: MR usability→SF	0.328 **	2.842	0.005	Supported	2.873
H2b: MR usability→PE	0.426 ***	3.796	0.000	Supported	2.873
H2c: MR usability→AT	0.586 ***	5.712	0.000	Supported	2.873
H2d: MR usability→CI	0.019	0.159	0.874	Not supported	3.211
H3a: SF→CI	0.292 **	2.945	0.003	Supported	2.497
H3b: PE→CI	0.295 *	2.472	0.013	Supported	2.347
H3c: AT→CI	0.290 **	2.885	0.004	Supported	2.647
H4a: MR features→SF→CI	0.115 *	2.117	0.034	Supported
H4b: MR features→PE→CI	0.097	1.670	0.095	Not supported
H4c: MR features→AT→CI	0.053	1.219	0.223	Not supported
H5a: MR usability→SF→CI	0.109 *	2.033	0.042	Supported
H5b: MR usability→PE→CI	0.126 *	1.967	0.049	Supported
H5c: MR usability→AT→CI	0.170 *	2.395	0.017	Supported

Note: * p < 0.05, ** p < 0.01, *** p < 0.001, 10,000 sample bootstrap. Standardized path coefficients (β) are reported, with significance levels indicated by asterisks. SF = satisfaction; PE = perceived enjoyment; AT = attitude; CI = continuance intention; VIF = variance inflation factor.

.

4.1. Common Method Bias

CMB refers to a systematic error that can arise when data for different variables are collected using the same method or from the same source, potentially distorting the relationships between those variables. CMB is viewed as a threat to distort the relationships between variables, leading to inflated or misleading correlations and undermining the validity of the study’s results. To assess CMB, this study employed two methods. First, according to Kock [102], the research model is considered free from CMB when the inner variance inflation factor (VIF) values in the full collinearity test are below the 3.33 threshold. The VIF values in the inner model of this study range from 1.000 to 3.315, as shown in Table 4, indicating the absence of significant multicollinearity issues. Second, a common method variance (CMV) was incorporated into the research model as a marker variable, establishing relationships with all constructs to assess the potential presence of CMB, as recommended by Liang et al. [103]. The results of the CMV analysis, provided in the Supplementary Materials, demonstrate that the substantive factor loadings are both high and statistically significant, whereas the method factor loadings are low and nonsignificant. This pattern suggests that the data are not substantially influenced by common method bias, thereby supporting the validity of the model’s constructs without significant redundancy.

4.2. Measurement Model Assessment

In accordance with Hair Jr et al. [101], the measurement model was evaluated through three critical steps to assess construct reliability, convergent validity, and discriminant validity. First, construct reliability was examined by evaluating internal consistency using indicators such as Cronbach’s alpha and composite reliability (CR). As presented in Table 1, Cronbach’s alpha values for the constructs in this study ranged from 0.736 to 0.887, exceeding the 0.7 threshold, while the CR values ranged from 0.848 to 0.931, also exceeding 0.7. These values fall within the acceptable thresholds suggested by Hair et al. [104], indicating that the measurement model demonstrates satisfactory internal consistency and reliability.

Second, convergent validity was assessed using the average variance extracted (AVE), which measures the extent to which multiple indicators of a construct are correlated, thereby confirming that all indicators capture the same underlying concept. The AVE values ranged from 0.652 to 0.817, all exceeding the 0.5 threshold, as recommended by Hair et al. [104]. The results indicate satisfactory convergent validity, with each construct being effectively represented by its respective indicators.

Finally, discriminant validity was assessed using the Heterotrait–Monotrait Ratio (HTMT) to confirm that the constructs are distinct and not excessively correlated. As suggested by Benitez et al. [105], the model is considered to have adequate discriminant validity when the HTMT values are below HTMT_0.85 (for a more stringent criterion of 0.85) or below HTMT_0.90 (for a more lenient criterion of 0.90). As shown in Table 2, the HTMT values in this study are all below 0.85, with the exception of two values falling between 0.85 and 0.90, suggesting that the model satisfies the requirement for adequate discriminant validity.

4.3. Structural Model Assessment

The structural equation model was carried out to analyze the path coefficients and t-values through bootstrapping resampling techniques, which provide robust estimates of standard errors and significance levels for the hypothesized relationships. As indicated in Table 4, KV-based MR features have a positive direct impact on SF ($\beta =0.393$, $p<0.01$), PE ($\beta =0.375$, $p<0.01$), and MR usability ($\beta =0.807$, $p<0.001$), but no significant direct effect on AT ($\beta =0.182$, n.s.) or CI ($\beta =0.007$, n.s.). Hence, H1a, H1b, and H1d are supported, except for H1c and H1e. KV-based MR usability has a positive direct impact on SF ($\beta =0.328$, $p<0.01$), PE ($\beta =0.426$, $p<0.001$), and AT ($\beta =0.586$, $p<0.001$), but no significant direct effect on CI ($\beta =0.019$, n.s.). Therefore, H2a, H2b, and H2c are supported, except for H2d. Regarding the influence of UX on STEM continuance intention in MRLEs, SF ($\beta =0.292$, $p<0.01$), PE ($\beta =0.295$, $p<0.05$), and AT ($\beta =0.290$, $p<0.01$) all have a positive impact on CI, as indicated by the significant path coefficients. Consequently, H3a, H3b, and H3c are supported. The percentages of variance explained ($R^2$ values) provide insight into the model’s fit and the extent to which the independent variables collectively account for the variability in the dependent variables, offering evidence of the model’s predictive validity. As shown in Table 4 and Figure 7, the research model explains 65.2% of the variance for MR usability, 53.4% for SF, 51.5% for PE, 54.9% for AT, 63.2% for CI.

4.4. Mediation Analysis

To analyze the mediating role of the organism component within the S-O-R paradigm in this study, UX factors, including SF, PE, and AT, are treated as mediators between the stimulus (MR features and usability) and the response variable (CI). This mediation analysis allows for a deeper understanding of how MR features and usability influence CI indirectly through their impact on user experience. As presented in Table 3, SF fully mediates the relationship between MR features and CI, as the direct effect is nonsignificant while the indirect effect through SF remains significant. Additionally, the results indicate that the direct effect of MR usability on CI is nonsignificant, while the indirect paths through SF, PE, and AT remain significant. This suggests that SF, PE, and AT fully mediate the relationship between MR usability and CI, demonstrating a full mediation effect.

5. Discussion

Grounded in the S-O-R paradigm, this study empirically evaluates the proposed research model in terms of the factors influencing STEM continuance intention within MRLEs. The research findings indicate that the proposed hypotheses (H1a, H1b, H1d, H2a, H2b, H2c) regarding the relationship between MR system stimuli and user experience were largely confirmed, with the exception of the nonsignificant relationship between MR features and attitude (H1c). These results strongly support the S-O relationship proposed in the S-O-R framework, demonstrating that MR features associated with knowledge visualization (e.g., 3D graphics, interface design, and operational functions) significantly affect perceived usability (e.g., PUCM, PU, and PEU). Furthermore, both MR features and perceived usability collectively exert a positive influence on UX, aligning with previous studies [41,106], which emphasize that the integration of well-designed technological features and optimized usability plays a pivotal role in shaping users’ overall experiences. This finding underscores the significance of knowledge visualization in the context of immersive learning environments. Specifically, the evaluation of the research model reveals the insignificant relationship between MR features and attitude. One possible explanation is that the relationship between KV-based MR features and learning attitude may be mediated by other factors, as verified by the mediation analysis through the path MR features→Usability→AT ($\beta =0.473$, $p<0.001$). This finding suggests that while the direct path from MR features to learning attitude is not significant, there is a full mediation effect where MR features influence learning attitude entirely through usability, highlighting the importance of optimizing usability to enhance learning attitude when using concept map tools as the form of knowledge visualization in immersive environments.

Results from the PLS analysis indicate that learners’ continuance intention significantly improves when they have a positive experience with the KV-based MR system (H3a, H3b, H3c). Specifically, learners’ intention to continue using the MR system is notably enhanced when they perceive a high level of satisfaction, a finding that is consistent with prior research on immersive system evaluations [107,108]. Yang et al. [109] underscores the critical importance of user satisfaction in maintaining engagement with immersive technologies. The current study provides evidence that this finding is equally applicable to using concept maps as the UI in MRLEs. This highlights the potential of concept maps not only as an effective means of knowledge visualization but also as a tool that can enhance user satisfaction, thereby ensuring sustained learning and interaction with the system. The perceived enjoyment, as an indicator of a pleasurable MR learning experience, has been shown in this study to be positively associated with continuance intention, which is in harmony with previous research [88,110]. Substantial empirical studies categorically confirm enjoyable experience as a crucial determinant of intention across a wide range of information systems [111,112]. The evidence retrieved from the present study adds to the literature by demonstrating that the well-designed KV-based features, centered around knowledge visualization, provided learners with a highly enjoyable learning experience, thereby fostering a positive response in their intention to continue using the MR system. The significant relationship between learning attitude and continuance intention indicates that a more positive attitude towards MR learning leads to a greater intention to adopt mobile MR technology, which is consistent with previous research [10,81]. Furthermore, the current study reveals deeper insights into the factors driving positive attitudes through both a direct effect of Usability on Attitude and an indirect effect through the pathway MR features → Usability → AT. This indicates that the MR features, as a core aspect of the KV design, play a crucial role in shaping learners’ attitudes, not only by directly enhancing usability but also by amplifying the effect of usability on learners’ attitudes.

This study inspects the underlying mechanisms of the organism constructs as mediators between stimulus constructs and response constructs. Interestingly, the full mediating effects (H4a) between MR features and continuance intention, as well as between MR usability and continuance intention (H5a, H5b, H5c), are revealed. These findings contribute to the field by enhancing our understanding of the critical role of UX in both capturing the impact of stimuli and driving the response of behavioral intentions within the S-O-R paradigm. The mediating effects of the organism in the S-O-R model are consistent with findings from previous studies [42,113]. Scholars have emphasized the importance of considering the mediating roles of organism factors, such as affective response, entertainment experience, and aesthetic experience, in the practical application of immersive technology [114,115]. This study bridges the gap between KV-based MR features, usability, and STEM continuance intention through the mediators of satisfaction, perceived enjoyment, and attitude, unraveling the motivating factors that drive learners’ desire to continue using MR for learning scientific knowledge.

5.1. Theoretical Implication

The research presented in this study makes several profound theoretical contributions. First, To the best of our knowledge, this study is the first research to employ concept maps as a knowledge-visualization tool integrated directly into the MR UI, enriching the current understanding of the pedagogical affordances of knowledge visualization and the technological effects of MR. It also responds to prior research calls [23,62,70] to explore the role of concept maps as scaffolding tools in immersive learning and their influence on STEM continuance intention. Although knowledge visualization, facilitated by interactive graphics to create, integrate, and apply knowledge, has garnered significant scholarly attention, researchers have yet to explore how learners’ perceptual experiences influence their intention to use such technologies, especially in immersive environments. The verification of H1 and H2 highlights the unique affordances of concept maps as a KV tool within MR environments and underscores the critical role of usability in enhancing the overall learning experience. The evaluation results of the MR features, including 3D graphics, interface design, and operational functions, provide a constructivist view of how MR-based visualization supports cognitive processing, aiding learners in organizing and integrating scientific knowledge more effectively. Recent studies suggest that concept maps, as effective knowledge-visualization methods for creating and transferring knowledge, enhance the effectiveness and timeliness of decision-making processes, potentially improving efficiency and increasing user satisfaction [116,117]. This study examines the effectiveness of concept maps as tools for KV in MRLEs in shaping learners’ affective experiences and ultimately influencing the sustainability of STEM education.

Second, this study extends CVTAE theory [30] into the domain of knowledge visualization in MR contexts. The findings also emphasize the importance of sustainable learning practices in MR environments by fostering positive UX in STEM education. The empirical evidence demonstrates that KV-based MR systems exhibit considerable efficacy in enhancing the learning experience within interactive multimodal environments. Specifically, the utilization of interactive concept maps as a knowledge-visualization tool reduces extraneous load by structuring information clearly, thereby optimizing learning in immersive educational settings. The proposed model integrates three critical dimensions of user experience, namely satisfaction, enjoyment, and attitude, and elucidates their interrelationships under the stimulus of immersive learning environments. Drawing on the CVTAE, the model posits that when an immersive learning activity is perceived as valuable and controllable, it instigates positive achievement emotions. Additionally, grounded in the TAM [66,118], the present study provides empirical evidence that positive sentiment plays a crucial role in shaping learners’ intention to continue using the KV-based MR system for sustainable STEM learning. These findings contribute to the literature by extending the conventional TAM framework, emphasizing positive sentiment as a psychological enabler that mediates the relationship between system design and behavioral intention.

Third, from a methodological perspective, this study employs PLS-SEM, a robust and sophisticated analytical technique, enabling a comprehensive examination of the relationships between KV-based MR features, user experience, and STEM continuance intention. PLS-SEM is particularly well-suited for analyzing complex models with multiple latent constructs, allowing for the simultaneous assessment of both direct and indirect effects. In this study, we specifically investigate the mediating role of user experience factors, such as satisfaction, enjoyment, and attitude, between system design and sustainable learning intention. Through the application of PLS-SEM, we offer an in-depth understanding of how immersive MR learning environments shape learners’ cognitive and affective responses, ultimately influencing their behavioral intentions. This methodological approach not only advances existing research on MR-enhanced learning but also provides valuable insights for educators and system designers seeking to optimize immersive learning technologies for the sustainability of STEM education.

5.2. Practical Implication

The current study offers ample evidence regarding the S-O-R model, providing guidance to educators and practitioners on designing appropriate interactive 3D knowledge visualization in virtual learning environments to meet specific teaching and learning requirements. This study emphasizes the pivotal role of KV-based MR design in enhancing the learning experience, particularly within immersive educational environments. With the growing importance of digital learning technologies and the increasing demand for effective knowledge-visualization tools, MR-based systems enable a new level of interactivity and engagement. By integrating concept maps as dynamic visual tools, the MR system facilitates the seamless transmission of complex scientific knowledge, offering learners an intuitive, immersive experience that fosters deeper understanding and retention. This research contributes to the literature by demonstrating how well-designed MR systems can transform the educational process, helping learners navigate and internalize STEM content more effectively. Furthermore, our findings suggest that educators should integrate KV-based MR systems into STEM curricula by designing sociability architectures where concept maps serve as the central tool for organizing and visualizing knowledge. These architectures, such as the game-based approach outlined in this paper, can incorporate interactive features and educational game mechanisms, which enable students to actively engage with the material and develop a more enjoyable and meaningful understanding of the content. Instructors should also consider incorporating testing features, such as the knowledge-check-based question-answering module adopted in this paper, within MR systems. These features can be tailored to control the learning journey, increasing student satisfaction upon completing the knowledge checks.

On the other hand, untangling the influence of hedonic aspects through UX versus utilitarian factors through KV-based MR usability is crucial for understanding their roles in the adoption and sustained use of MR technologies. In the proposed model, system usability plays a direct role in shaping UX, enhancing the hedonic experience by improving satisfaction and enjoyment. Meanwhile, MR features influence usability, which in turn impacts user attitudes, highlighting the importance of usability in fostering both utilitarian and hedonic aspects of the learning experience. We recommend that immersive learning system designers and course instructors ensure the integration of reliable and stable MR systems, focusing on factors such as system stability, real-time tracking accuracy, and minimal latency to avoid technical disruptions during the learning process. Additionally, MR systems should be designed with robust tracking capabilities to ensure accurate and responsive interaction with the virtual environment, which can significantly impact user engagement and overall satisfaction. By focusing on both usability and system reliability, educators can create immersive learning environments that support a seamless, attractive, and effective learning experience. This research provides valuable insights into how the interplay between hedonic factors (manifested through UX) and utilitarian factors (manifested through usability) influences learners’ decisions towards MR system usage, underscoring the need for a balanced integration of both dimensions to encourage sustained usage and ongoing willingness. MR developers and instructors should stimulate balanced scaffolding functions that provide learners with high levels of support, enhancing both the hedonic experience and the utilitarian aspects to foster a strong sense of participation.

5.3. Limitations and Directions for Future Research

Although the findings in this study offer a rigorous picture of the mechanisms underlying learners’ perceptions and adoption of concept maps as a form of knowledge visualization in MR learning contexts, several limitations remain that need to be addressed in future research.

First, the sample in this study was limited to high school students in China enrolled in physics courses. Future research may aim to broaden the sample to include participants from diverse geographical and cultural backgrounds spanning various age groups. Moreover, it is essential to assess the effectiveness of the KV-based MR system across other STEM disciplines, such as mathematics, chemistry, and biology, to evaluate its broader applicability and generalizability in different educational contexts.

Second, the proposed S-O-R model primarily examines how KV-related MR features and system usability influence STEM continuance intention through the mediating role of user experience (UX). The model accounts for 63% of the variance in participants’ willingness to continue adopting MR in their future studies. However, the remaining unexplained variance may stem from factors not included in the model, such as individual differences in prior technology experience, motivational traits, or external environmental influences. These factors remain to be explored in order to fully understand the impact on the adoption and sustained use of MR technologies in educational contexts.

Third, the data in this study primarily rely on participants’ subjective perceptions and self-reported questionnaires. Future research could benefit from incorporating more objective data, such as observational data, to mitigate the potential biases introduced by individuals’ perceptions.

Fourth, the data in this study were collected based on learners’ perceptions after a relatively short trial period. Future studies should consider longitudinal tracking to assess the long-term effects and the sustainability of MR technology adoption and its impact on learning effects.

6. Conclusions

This study explores the potential pedagogical affordances of using concept maps as scaffolding tools for knowledge visualization in MR learning environments. The findings reveal that user experience is significantly influenced by KV-based MR features and usability, strengthening the link between MR system design and the sustainability of STEM education [10]. While prior research has highlighted the benefits and challenges of knowledge visualization in digital multimedia, the proposed S-O-R model makes a notable contribution by demonstrating that user experience is a central factor in shaping learners’ willingness to adopt concept maps as a user interface in MR learning environments, supporting the sustainable adoption of immersive learning technologies. This study extends the existing body of knowledge by providing empirical evidence on how the interplay between hedonic and utilitarian factors affects learners’ decisions to engage with MR-based knowledge-visualization systems. Practical implications are offered for educators and practitioners, encouraging the implementation of more effective knowledge-visualization strategies to enhance the sustainability of STEM education and support positive learning experiences in immersive educational settings.