Christos Troussas

Mixed reality (MR) presents potential to provide personalized learning through alignment of users’ activity with instructional assistance. Still, the nature of such adaptations and the educational value of personalization in this context have not been fully explored yet. This systematic review synthesizes 40 publication reports retrieved from Scopus (2016–2026) on personalized mixed reality in higher and professional education focusing on the aspects of educational context, technology configuration, means of adaptation, and learning outcomes. This review explores how individual performance, movements, gaze and dialogue affect the adjustment of feedback, coaching and practice. Five categories of personalized MR were revealed: coaching and rehearsing based on performance, feedback driven by sensors, eye-tracking-based attention management, guidance by an expert remotely connected through mixed reality and conversational or embodied artificial intelligence. Human coaching is prevalent among the evidence base, especially in teacher education, whereas recent research applies increasingly complex technology-assisted interventions in medicine, technology and other branches of higher education. The clearest reported benefits concern proximal teaching behaviors and procedural performance. Evidence for sustained unaided learning and authentic transfer remains limited. Bundled interventions require the careful attribution of outcomes to their individual components. The process of personalization was done mostly while performing current tasks or during sessions. There is no evidence available on persistent computational learner models across multiple sessions. This review summarizes current practices of personalized support in mixed reality and identifies a further research agenda of adaptive-learning-pathway design and evaluation.

Social and semantic data have become one of the main raw materials of contemporary artificial intelligence (AI)

Collaborative learning scripts are mainly fixed, and collaboration dashboards do not directly trigger teachers to intervene. We thus introduce a hybrid intelligent decision-support system and a simulation benchmark with introduced group dysfunctions, known onsets and a true per-group oracle. The system includes quantile-anchored fuzzy inference (35 rules), a lexical content classifier, calibrated gradient-boosting detectors with 3-levels explanations, expected-benefit selection from a closed operator catalogue, and NSGA-II regrouping. Teacher approval is designed, but was not evaluated. After 20 simulated datasets, the detectors achieved macro-F1=0.591, macro-AUROC=0.870 and isotonic calibration reduced expected calibration error to 0.018-0.025. The fuzzy indicators were practically indistinguishable from no indicators in accuracy (posterior probability of a difference within ±0.01: 0.96); the intended value of traceable rule-level evidence is yet to be tested with teachers. Generalisation was main limitation: under changed mechanism, macro-F1 fell from 0.586 to 0.297; in a held-out topic, the full model fell below the interaction-only model (0.402 versus 0.437); in new paraphrases, misconception F1 dropped to 0.146. SHAP explanations came close to occlusion and interventional SHAP explanations and greatly surpassed a model-free ranking, but the detector dependent on construct irrelevant features so verbalisations had to be post-filtered by a consistency predicate. At the operating point determined on training data, operator selection was 0.399 of the oracle's harm-mitigation with 18.7% undesired interventions, matching a fixed operator map. Class-level regrouping increased simulated harm; greedy-seeded NSGA-II was only slightly better than other greedy-seeded searches. These simulation results must be regarded as upper bounds.

Large language models can produce remediation suggestions for IT incidents, but the evidence behind their suggestions and the technical correctness of their suggestions are unclear. We evaluate a prototype escalation-aware retrieval-augmented generation (RAG) system for supervised first-line incident triage, using IBM TechQA retrieval, a QLoRA specialized Llama-3-8B-Instruct model, explicit abstention, and Jira ticket escalation. The final reranker and decision thresholds were trained or calibrated on TechQA training questions, but previous choices of retrieval models and contexts used to train the adapter depended indirectly on development questions, so the overall development process is not fully leakage-free. On 205 missing-evidence TEST cases, correct abstention increased from 8.5% (Base) to 42.4% (QLoRA), and hazardous answering decreased from 91.5% to 57.6%. The gate had higher correct abstention (53.7% vs. 42.4%; a significant 11.2 percentage-point advantage, Holm p = 0.013) with no detectable difference in false abstention when operating at their own training-calibrated operating points. At retrospectively matched false abstention, QLoRA had an estimated advantage of +11.2 points over the gate. This different, TEST-threshold-selected contrast was exploratory and not statistically significant. In exploratory NLI analysis, 26.1% of citation-valid QLoRA answers were flagged as containing an unsupported step. NLI–human agreement did not meet the pre-specified acceptance criterion (κ = 0.324 < 0.5), so these estimates cannot be interpreted as validated semantic-grounding rates. In the paired Jira simulation, 64.1% of tickets were closed with a proposed resolution, including 54.9% of naturally unanswerable tickets. In an internal assessment by two research-team annotators, an estimated 14.2% of all incidents were automatically closed with fully correct proposed answers (not independently verified repairs) and potentially risky actions appeared in 24.0% (6/25) of sampled automatically closed missing-evidence tickets. Results support supervised triage and human escalation, not safe autonomous technical remediation.

Intelligent sensor technologies now provide unprecedented multimodal access to human affective and cognitive processes, spanning physiological, ocular, facial, kinematic, ambient, and behavioral streams. Yet, despite dramatic advances in acquisition and deep representation learning, the pipeline from raw signals to psychologically meaningful, actionable understanding remains fragile. Deep-only architectures excel at pattern extraction but struggle with contextual reasoning, uncertainty communication, and human-facing explanation; symbolic-only frameworks resist noisy, high-dimensional streams. This perspective argues that the field’s next inflection point lies not in richer sensors alone but in the interpretive layer that turns signals into states. We propose an integrative view in which neuro-symbolic fusion couples continuous sensor evidence to psychologically grounded symbolic primitives, LLMs act as auditable semantic reasoners rather than opaque classifiers, and explainability is treated as a design constraint rather than a post hoc addition. Synthesizing a decade of work on affective computing, fuzzy learner modeling, neuro-adaptive multimodal interaction, and explainable human–AI collaboration, we motivate a research agenda organized around grounded representations, uncertainty-calibrated LLM reasoning, and reflexive explanation. The aim is a class of sensing systems whose intelligence is measured not by accuracy alone but by the quality of the states they help humans understand.

Artificial Intelligence (AI)-based educational systems increasingly support personalised learning, adaptive feedback, and learning analytics. However, these capabilities are often addressed separately, with comparatively less attention paid to their integration into interpretable, teacher-facing decision-support frameworks. This study presents the e-Teacher Assistant, a human-centred educational framework that integrates xAPI-style Learning Record Store (LRS) analytics, dynamic learner modelling, Sugeno-type fuzzy inference using interpretable IF–THEN rules, adaptive learning support, and a dual Student Model–Teacher Model architecture. The framework was evaluated during a three-month authentic deployment in a vocational education Computer Networks course involving 117 learners and four educators. Learner evaluation combined a structured questionnaire, open-ended responses, and LRS-based behavioural analytics, while educators evaluated the Teacher Model through a questionnaire and qualitative responses. Learners reported predominantly positive perceptions of the system, including ease of use (92.3%), usefulness of feedback (95.7%), helpful interaction with the AI Assistant (94.9%), support for independent learning (88.0%), and support for educators through learning analytics (87.2–88.0%). Fairness and objectivity were also evaluated positively by 86.3% of learners. All four educators evaluated the Teacher Model positively, and all strongly agreed on its overall usefulness and its support for progress monitoring. Prior familiarity with digital learning tools was significantly associated with evaluations across all seven questionnaire dimensions. LRS analytics further documented learner engagement, repeated assessment activity, and contextual AI use; AI-assisted examination support was recorded in 64.42% of learner–chapter records. Overall, the findings provide initial empirical support for the feasibility of integrating interpretable learner modelling, adaptive AI-assisted support, LRS-based learning analytics, and teacher-facing decision support in an authentic vocational education setting while preserving educator oversight and pedagogical responsibility.

Assessment of VR experience quality has long been performed through measurement tools that are focused on measuring presence, usability or simulator sickness or embodiment separately. This approach becomes less and less applicable to the modern standalone HMDs where users interact using both handheld controllers and optical hand tracking. Not only is the sense of location in a virtual environment an important aspect of modern VR experience, but there is also a sense of ownership of one’s body in this environment, agency, naturalness of actions, tracking precision, Cognitive Fluency and comfort. This study introduces a multidimensional questionnaire for the assessment of embodied and modality-dependent interaction in modern VR—the Embodied Presence and Interaction Scale for Virtual Reality (EPIS-VR). The paper provides a synthesis of relevant literature and describes a structure of the proposed 28-item scale based on seven dimensions: Spatial Presence, Embodied Ownership, Embodied Agency, Interaction Naturalness, Interaction Fidelity, Cognitive Fluency and Interaction Comfort. Empirical validation includes sample descriptions, factor-analytic findings, reliability scores, convergence and discrimination validity tests, and comparisons of handheld and optical hand-tracking modalities. The EPIS-VR provides an initially validated multidimensional evaluation model for controller-based and hand-tracking interaction within an educational informatics context. Further studies are required to establish its applicability across other populations, VR domains, interaction modalities, and hardware systems.

This study highlights the emerging need for enhanced user interaction in virtual reality (VR) systems designed for cultural heritage applications. Conventional VR systems often struggle to adapt to the diverse range of users, particularly in considering the various levels of technical skills and familiarity with VR technologies, as well as cultural knowledge, thereby limiting their ability to deliver content and engage users effectively. To address these gaps, the proposed fuzzy-based personalized VR system employs fuzzy weights to dynamically evaluate both users' Computer/VR skills and their understanding of cultural heritage. This dual-layer fuzzy-based approach enables VR system to adapt its functionality by presenting content through UI elements tailored to the users’ combined technical and cultural level of expertise. The combination of computer/VR skills and cultural knowledge ensures that both the how (technical interaction) and the what (cultural content) of the VR experience are personalized, enabling users to engage with the virtual environment effectively and efficiently. The VR system was evaluated using a T-Test on an experimental and a control group to assess its effectiveness in interface and content personalization compared to a conventional VR system. The findings indicate that the proposed VR system significantly outperforms the conventional approach, providing a more immersive, adaptive, and engaging user experience.

This study presents an adaptive learning framework that integrates fuzzy logic and learning analytics to support personalized education and multi-factor student assessment. The proposed system combines cognitive and behavioral indicators to provide an interpretable representation of the learner’s state within a dynamic digital learning environment. The architecture is based on adaptive learner modeling and classroom-level monitoring mechanisms, enabling personalized guidance, adaptive content sequencing, and continuous performance monitoring at both individual and classroom levels. A core contribution of the approach is a fuzzy logic-based evaluation mechanism that aggregates multiple signals, including quiz performance, time spent on theory, help-seeking behavior, and system interaction patterns. These inputs are transformed into fuzzy sets and combined through inference rules to produce interpretable learning level estimates aligned with Bloom’s taxonomy. The approach is grounded in Vygotsky’s Zone of Proximal Development, supporting adaptive scaffolding and targeted instructional interventions. The evaluation results demonstrate a strong correlation between the model outputs and conventional exam performance (r ≈ 0.91), while exhibiting reduced variability (SD ≈ 0.15 compared to SD ≈ 0.20), indicating a more stable representation of learner performance. Furthermore, statistical analysis confirms that the differences between traditional and model-based scores are significant (p < 0.01), suggesting that the proposed approach captures additional dimensions of learner behavior beyond conventional grading metrics. Overall, the findings indicate that integrating fuzzy reasoning with behavioral analytics enables a more interpretable, stable, and pedagogically grounded approach to learner assessment, supporting adaptive and interpretable personalized learning.