Product neural network is one of the most popular deep learning models, because of its good feature extraction ability and classification prediction effect, it is widely used in the field of image recognition, face recognition and natural language processing, etc. The CNN’s local connectivity, weight sharing and downsampling are the significant features that differentiate it from other network models. User emotion recognition is the basis of human-machine emotional interaction, for in the welcoming scene, the user emotion performance in the interaction process is mostly composite emotions, so the recognition of the basic emotion category can not satisfy the robot for the characterization of any user emotion, for this kind of problem, this paper proposes an emotion cognition method based on the characterization of facial features, and under the role of convolutional neural network, the human facial emotion recognition is carried out. research and exploration. Combined with the research of human facial emotion recognition under big data technology is a new field of big data used in human research, this study is based on the research foundation of previous scholars, and carries out in-depth exploration and analysis, which plays a certain role in promoting the research of human facial emotion recognition.

Emotion-recognition-enabled smart care devices are increasingly proposed for home-based rehabilitation and elderly care, yet evidence on prospective user and family caregiver acceptance remains limited. This study investigated public attitudes, concerns, interaction preferences, and design expectations regarding such devices through a cross-sectional online survey of 506 adults in China. Respondents answered from either a prospective individual-user perspective ( n = 339) or a family caregiver perspective ( n = 167). Descriptive statistics, Top-2 Box summaries, and visual comparisons were used to analyze acceptance, scenario priorities, sensing preferences, data-sharing attitudes, and perceived device roles. Overall, respondents showed favorable but conditional acceptance. Privacy transparency, reliability, and recognition accuracy emerged as key adoption conditions, while concerns about emotional data misuse and machine accuracy remained salient. Participants prioritized home-based rehabilitation and health management scenarios, preferred wearable physiological sensing over camera-based monitoring, and favored human-in-the-loop interaction over fully automated responses. Family caregivers showed descriptively stronger endorsement than prospective individual users on several selected items, although the overall pattern was broadly consistent across groups and should not be interpreted as evidence of statistically significant subgroup differences. The findings suggest that emotion-recognition-enabled smart care devices may be more acceptable when designed around four principles: safety first, human-in-the-loop control, low-intrusion sensing, and bounded empathy. Because the sample consisted mainly of online adult respondents rather than actual older adult users, the findings should be interpreted as exploratory evidence on prospective acceptance rather than as direct evidence of real-world adoption.

This study examines how university students engage with generative AI (GenAI) across instrumental and relational domains using two independent, domain-specific exploratory surveys of Chinese university students (Study 1: N = 517; Study 2: N = 508). Study 1 assessed instrumental indicators, including agency, delegation/offloading, trust/adoption, and perceived gains, whereas Study 2 assessed relational indicators, including anthropomorphism, emotional safety/control, disclosure/reliance, and displacement/withdrawal. Latent profile analysis identified two profiles in each study. In Study 1, “Augmented Users” showed higher endorsement of agency, trust, gains, and delegation than “Limited Users.” In Study 2, “Attached Users” showed higher endorsement of anthropomorphism, emotional safety, disclosure, and displacement/withdrawal than “Uninvolved Users,” and also reported higher loneliness and slightly lower social satisfaction. Because the two studies used independent samples and different indicators, the findings should be interpreted as separate domain-specific profile analyses rather than a direct test of a unified person-level typology. The study provides exploratory person-centered evidence for differentiating instrumental and relational patterns of GenAI engagement in higher education.

1. Immersive VR is increasingly used in language learning because it can combine spatial presence, multimodal input, embodied interaction, and situated visual context. These affordances are relevan...

BACKGROUND Background: Immersive exercise technologies are increasingly proposed to support healthy aging, yet sustained adoption among older adults remains inconsistent. Prior work often treats acceptance as a static decision and frequently aggregates virtual reality (VR) and mixed reality (MR), potentially obscuring modality-specific adoption mechanisms. OBJECTIVE Objective: This study aimed to examine how older adults’ adoption of immersive Tai Chi exercise evolves over time and whether VR and MR support sustained engagement through different mechanisms. Specifically, we investigated phase-specific barriers and facilitators across entry, adaptation, and adoption negotiation, as well as differences in trust development and continued use between VR and MR conditions. METHODS Methods: Guided by the Dynamic Barrier Framework (DBF), we conducted a qualitative-dominant mixed-methods longitudinal study of older adults completing repeated immersive Tai Chi training. Participants completed one initial session and six repeat sessions. Semi-structured interviews captured evolving barriers and facilitators across DBF phases (entry, adaptation, and adoption negotiation). Quantitative measures of subjective experience and cybersickness, together with session-level performance logs, were integrated for descriptive triangulation. RESULTS Results: Thirty-five participants were enrolled (VR: n = 16; MR: n = 19); 34 provided complete quantitative data (VR: n = 16; MR: n = 18). Across modalities, adoption unfolded through three sequential phases with systematic shifts in barriers and facilitators. Early resistance was driven more by uncertainty and perceived control than by physiological discomfort alone. VR engagement was more closely linked to immersive and affective value, whereas MR engagement emphasized skill adaptation and the emergence of trust over time. Performance logs indicated learning across repeated sessions, aligning with self-reported experience. CONCLUSIONS Conclusions: Older adults’ immersive exercise adoption is dynamic and phase-sensitive. VR and MR may foster sustained engagement through differentiated mechanisms, suggesting the need for phase-tailored support strategies in community deployment. CLINICALTRIAL Trial Registration: Not applicable.

Background: Immersive virtual reality (VR) is increasingly used for embodied training, but its practical value for older adults depends on whether the experiential benefits can be achieved without unacceptable discomfort or overstated claims about motor learning. Objective: This randomized study compared head-mounted immersive VR with standardized, content-matched 2D video instruction for Tai Chi movement-encoding in community-dwelling older adults. Methods: A total of 106 older adults were randomized to immersive VR (n=56) or video instruction (n=50). The primary analysis included 92 participants with complete age, pretest, and posttest data (VR: n=45; Video: n=47). The primary outcome was a 5-item movement-encoding knowledge score (range 0‐20), analyzed using analysis of covariance, with posttest score as the dependent variable and group, pretest score, and age as predictors. Secondary outcomes were intrinsic motivation, game experience, and learning engagement; safety, tolerability, and feasibility were assessed using the Intrinsic Motivation Inventory, Game Experience Questionnaire, Utrecht Work Engagement Scale–Short Form, participant flow, and Virtual Reality Sickness Questionnaire. VR system telemetry was analyzed descriptively as an exploratory process outcome. Results: Adjusted posttest movement-encoding performance did not differ between VR and video instruction (adjusted difference −0.27, 95% CI −2.25 to 1.72; P =.79; partial η ² =0.001). The Game Experience Questionnaire results favored VR for competence (Video: mean 3.36, SD 1.21; VR: mean 4.15, SD 0.92; q =0.009), flow (Video: mean 3.77, SD 1.26; VR: mean 4.46, SD 0.76; q =0.014), and lower tension (Video: mean 1.76, SD 1.04; VR: mean 1.27, SD 0.63; q =0.037), whereas challenge did not remain significant after false discovery rate correction ( q =0.073). In the VR Group, recognition accuracy was associated with posttest movement-encoding score (n=44; r =0.317, P =.04), but pose accuracy was not (n=44; r =0.129, P =.41). The Virtual Reality Sickness Questionnaire scores were low among valid VR participants (total score mean 3.77, SD 11.12), and 36 out of 46 (78%) participants reported zero symptoms. Conclusions: With a strong active control, immersive VR did not produce higher short-term Tai Chi movement-encoding performance than video instruction. Among participants who completed the VR session, the immediate game-experience profile was more favorable. These findings indicate that VR may offer experiential benefits, while also highlighting the need to manage cybersickness and technical exclusions and to distinguish movement knowledge from actual motor execution.