• HARP metrics quantify diverse Q-C relationships across ten water constituents. • Land use and local mechanisms influence Q-C hysteresis during storm events. • Biased sampling and strong hysteresis can significantly affect load accuracy. • HARP-based mechanisms help reduce uncertainty in storm-driven load estimates. Understanding flow (Q)-concentration (C) interactions is essential for providing effective catchment management strategies to reduce loads of contaminants from land to waterways. We tested if a non-linear Hysteresis Area, Residual, and Peak (HARP) analysis tool could improve Q-C relationships in streams and provide superior nutrient load estimates during storm events compared with linear or univariate models. Q-C patterns and load estimates were tested with data from three stations in a subtropical stream in South East Queensland, Australia, from 2007 to 2022. HARP revealed distinct site- and constituent-specific characteristics associated with hysteresis. Clockwise hysteresis patterns predominated at two stations where the subcatchment was dominated by intensive cropping, reflecting rapid nutrient mobilization during rising flows, while the station where subcatchment had greater forested area showed anticlockwise patterns, indicating a delayed nutrient response to rising flow. Strong inter-station differences in HARP parameters for nitrate-nitrogen (NO 3 -N) were likely due to stream interactions with groundwater, while ammonium-nitrogen (NH 4 -N) and phosphate-phosphorus (PO 4 -P) showed no significant differences among stations. Substantial errors in sediment and nutrient load estimates may occur when hysteresis effects are overlooked during storm events, particularly where there is an asymmetric hysteresis pattern between the rising and falling limbs of the hydrograph. Load estimates depend on both the sampling strategy and calculation methodology, with inaccuracies compounded when sampling is unevenly distributed during storms with strong hysteresis, and modified by subcatchment land use and groundwater contributions.

• ZooMS study of 988 bones from the Palaeolithic site, Geißenklösterle Cave. • Carnivore damage three times more prevalent on indeterminate fossils than those with diagnostic features. • Fragment size influences species identifications in ZooMS studies. • Earliest anthropogenic modification on Rhinocerotidae remains in the Swabian Jura. Geißenklösterle Cave, a Palaeolithic site located in the Swabian Jura of southern Germany, has yielded multiple Middle Palaeolithic strata and some of the earliest evidence for the Aurignacian and Gravettian in Europe. Rich assemblages of three-dimensional figurines, personal ornaments, and musical instruments position Geißenklösterle Cave, and the complex of caves in the region, as significant in our understanding of the Palaeolithic in Europe. Extensive osseous assemblages preserved through the Palaeolithic layers of the site have allowed for thorough zooarchaeological research, aimed at reconstructing site use and human behavior. Drawing from this established research, we applied ZooMS (Zooarchaeology by Mass Spectrometry) to 988 morphologically non-identifiable bone fragments from Geißenklösterle Cave to develop strategies for the integration of non-diagnostic material into the previously established datasets. Our results indicate excellent preservation conditions for collagen, with a 98.4% success rate for ZooMS. The inclusion of taphonomic analyses in our study suggests predators played a significantly bigger role in bone fragmentation at the site than previously assumed, with 31% of bones included in this study exhibiting some kind of carnivore damage. Rare instances of anthropogenic modifications on bones were also identified, including the earliest instance of a cut-mark on a woolly rhinoceros ( Coelodonta antiquitatis ) bone in the Swabian Jura. Finally, we demonstrate the potential biases introduced through sampling strategies regarding the size of bone fragments analyzed in our study. Our results provide valuable insights for the discourse on both Palaeolithic archaeology as well as the best-practice integration of ZooMS into the zooarchaeological workflow.

Adaptive metamodeling approach for real-time taxiing time estimation on airport surface

3D MR image acquisition is inherently time intensive, rendering it susceptible to patient motion during scanning. This may introduce significant blurring and artifacts, potentially necessitating reacquisition. We propose a modular framework to retrospectively correct for intrascan motion in 3D brain MRI, without active motion tracking. Serving as the backbone of our approach is an existing distributed and incoherent sampling scheme (DISORDER), combined with a fast network trained for highly undersampled reconstruction. This enables approximate reconstructions of anatomy after every few seconds, using only a tiny fraction of k-space data (< 2%). While these reconstructions are only approximate, we postulate they are sufficient to estimate motion patterns at said temporal resolution. Groupwise registration, notable for its elimination of registration bias, is utilized for estimating rigid motion parameters, which are leveraged to reconstruct the measured data with reduced motion artifacts. The approach was evaluated on 94 retrospectively and 3 prospectively motion-corrupted invivo 3D T1-weighted brain MRI acquisitions. The estimated motion parameters matched the known retrospective motion with 0.06 mm and 0.13° accuracy, resulting in an improvement in reconstruction quality from to SSIM for the retrospective scans. The prospective scans improved from to SSIM after correction in the case of gradual motion and from to SSIM for extreme motion. In conclusion, the proposed approach, that is free of external tracking devices or navigators, successfully estimated and corrected 3D motion between small subportions of a scan. This resulted in vastly improved image quality, making volumetric MRI substantially more tolerant to motion.

Semi-supervised learning (SSL) has achieved notable progress in medical image segmentation. To achieve effective SSL, a model needs to be able to efficiently learn from limited labeled data and effectively exploit knowledge from abundant unlabeled data. Recent developments in visual foundation models, such as the Segment Anything Model (SAM), have demonstrated remarkable adaptability with improved sample efficiency. To seamlessly harness foundation models in SSL, we propose a SAM-driven cross prompting framework with adaptive sampling and prompt consistency for semi-supervised medical image segmentation, named CPAC-SAM. Our method employs SAM's unique prompt design and innovates a cross prompting strategy within a dual-branch framework to automatically generate prompts and supervision across two decoder branches, enabling effective learning from both scarce labeled and valuable unlabeled data. To ensure the quality of prompts for unlabeled data and provide meaningful supervision in the cross prompting scheme, we propose an innovative prototype-guided grid sampling strategy with adaptive intervals to simultaneously improve the reliability of the prompt selection area and ensure both adequate prompt density and complete target coverage. We further design a novel prompt consistency regularization to reduce SAM's prompt sensitivity and to enhance the output invariance under different prompts. We validate our method on five medical image segmentation tasks, encompassing both 2D and 3D scenarios. The extensive experiments with different labeled-data ratios and modalities demonstrate the superiority of our proposed method over the state-of-the-art SSL methods, with more than 4.1% and 3.8% Dice improvement on the breast cancer segmentation task and left atrium segmentation task, respectively. Our code is available at: https://github.com/JuzhengMiao/CPAC-SAM.

Pain is a common, complex, and debilitating non-motor symptom of Parkinson's disease that often requires a multidisciplinary and multimodal approach to care. However, little is known about how healthcare providers experience and navigate pain care provision in Parkinson's disease, particularly within the Australian healthcare context. This study aimed to explore the perspectives of Australian healthcare providers involved in the management of pain in people with Parkinson's disease. A qualitative descriptive research methodology was used to underpin this study. Healthcare providers involved in the management of pain in people with Parkinson's disease were recruited using a quota sampling strategy. Data were collected using semi-structured interviews and analyzed thematically. Nineteen healthcare providers participated, including five from nursing, five from neurology, five from physiotherapy, three from exercise physiology, and one from general practice. Analysis of the data resulted in six themes: (1) collaborating for holistic pain management; (2) determining etiology; (3) screening and managing the non-motor symptoms of Parkinson's disease that interact with pain and its treatment; (4) provider focus drives nuanced pain therapy; (5) involving carers in the assessment and treatment of pain; and (6) educating patients and healthcare providers to achieve effective pain management. Within these themes, participants highlighted the central role of Parkinson's nurse specialists in coordinating care, facilitating communication across disciplines, and supporting both patients and healthcare providers through education and guidance. This study offers the first in-depth exploration of healthcare providers' perspectives on pain care in Parkinson's disease within Australia. The findings highlight the need for a collaborative, multidisciplinary approach supported by targeted patient and healthcare provider education to enhance communication and continuity of care, with Parkinson's nurse specialists playing a pivotal role in coordinating services and serving as a resource across the care network. Strengthening these elements presents a clear pathway for advancing the provision of pain care services in people with Parkinson's disease. Standards for Reporting Qualitative Research (SRQR) guidelines.

Social-emotional competence plays a pivotal role in adolescent development. Currently, teacher-student relationships are widely recognized as a critical determinant of adolescents' social-emotional competence, yet the underlying mechanisms remain insufficiently explored. This study employed a multicenter longitudinal design to examine the associations and pathways linking teacher-student relationships and adolescents' social-emotional competence, with particular attention to the mediating roles of school climate and interpersonal trust. Using a multicenter, stratified cluster sampling strategy, a longitudinal survey was conducted between June and December 2024 with 793 high school students. The survey instruments assessed demographic characteristics, teacher-student relationships, social-emotional competence, school climate, and interpersonal trust, and the relationships among variables were analyzed through structural equation modeling. Findings indicated that adolescents' social-emotional competence was at a moderate level. Teacher-student relationships positively predicted social-emotional competence, with school climate serving as a mediator between teacher-student relationships and social-emotional competence, while interpersonal trust mediated the link between teacher-student relationships and school climate. Thus, teacher-student relationships were directly related to adolescents' social-emotional competence and also indirectly influenced it through a sequential mediation pathway involving interpersonal trust and school climate. These results underscore the importance of fostering positive teacher-student relationships, strengthening interpersonal trust, and cultivating a supportive school climate to promote adolescents' social-emotional competence. (PsycInfo Database Record (c) 2026 APA, all rights reserved).

Life beyond diagnosis: The psychosocial experiences of individuals with a sibling with autism: A qualitative study.

Validity of spot urine samples for estimating systemic fluoride exposure in children across diverse fluoridation modalities.

HCAR1 antagonist screening based on boundary-selected negative sampling strategy and multi-level graph neural network.

Causality-driven dual-domain network enhanced by Gaussian splatting for magnetic resonance image reconstruction.

Alzheimer's disease neuropathological changes (ADNC)-operationalized with semi-quantitative parameters-represent the consensus-based gold standard for diagnostic evaluation of disease severity. Although useful, ADNC diagnostic frameworks have limitations, particularly in advanced disease stages where pathological severity varies widely within a given diagnostic category. Further, some individuals lacking cognitive impairment are inappropriately categorized as having severe ADNC. In this study, quantitative pathology metrics and alternative tissue sampling schemes were integrated with data about premortem cognitive status, in order to derive clinically informed neuropathologic diagnostic thresholds. Specific goals of the current study were to generate data-driven, standardized diagnostic cut-points, with the most severe stages of ADNC having consistent implications: Braak neurofibrillary tangle (NFT) stage V cases being always impaired (MCI or demented) and Braak NFT stage VI cases being always demented. Utilizing whole-slide imaging and AI-based image analysis, object-based (NFT counts) and pixel-based (phosphorylated tau [pTau] burden) quantifications were compared across neocortical regions in three subsamples of cases from the University of Kentucky ADRC autopsy cohort (n = 329, all with clinical evaluations within 2 years of death). We also compared between HALO- and Aperio-based platform results, and between AT8 and PHF-1 pTau antibodies. Applying refined thresholds enabled reclassification of cases previously misaligned with their digitally determined appropriate status: 17% of cases were thus reclassified. The use of commercially available software, standardized classifier architectures, and interoperable analysis pipelines facilitated scalable and reproducible digital quantification. Cross-institutional validation at University of Texas San Antonio, with the same algorithms applied in both research centers, confirmed near-perfect agreement of pathology counts, underscoring shareability and the feasibility of harmonized digital workflows for collaborative research and diagnostic purposes. These findings support the integration of quantitative digital pathology into standard neuropathological protocols and provide a scalable model for future multi-site studies. Enabling comparisons of analytical platforms, pTau antibodies, and anatomical sampling strategies, an updated workflow demonstrated high reproducibility and consistent clinical-pathological correlations.

Fast food consumption is a growing public health concern, particularly among adolescents, due to its association with obesity and chronic diseases. Understanding the behavioral determinants of fast-food avoidance can inform the design of effective interventions targeting male adolescents. This study aimed to investigate the determinants of fast-food avoidance behaviors among male high school students in Shiraz, Iran, using the Theory of Planned Behavior (TPB) and Structural Equation Modeling (SEM). A cross-sectional study was conducted among 862 male students in grades 1 and 2 of high school in Shiraz in 2025. A multi-stage sampling strategy was applied, whereby 16 high schools were randomly selected from four educational districts. Within selected schools, classes and students were recruited using a convenience-based approach, and all eligible students present in the selected classes were invited to participate. Data were collected using a demographic questionnaire and a validated TPB-based instrument assessing knowledge, attitude, subjective norms, perceived behavioral control (PBC), behavioral intention, and fast-food avoidance behavior. Data analysis was performed using JAMOVI (v2.5.6), and SEM was employed to test the hypothesized TPB model. The final TPB model demonstrated excellent fit indices (χ²/df = 2.04, CFI = 0.94, RMSEA = 0.05). Attitude (β = 0.58), subjective norms (β = 0.34), and PBC (β = 0.26) significantly predicted behavioral intention (p < .001), which in turn strongly predicted fast-food avoidance behavior (β = 0.49, p < .001). PBC also had a direct effect on behavior (β = 0.31). Knowledge exerted an indirect effect on behavior through attitude and intention. Demographic variables, including parental education, household income, and maternal employment, were associated with TPB constructs. Fast-food avoidance behaviors among male adolescents are primarily influenced by motivational and control beliefs. From a practical and policy perspective, school-based nutrition education programs and adolescent health policies should prioritize strengthening positive attitudes, enhancing perceived behavioral control, and reinforcing supportive social norms to promote healthier dietary behaviors and reduce fast-food consumption among adolescents.

Gender-based violence remains a pervasive public health challenge in India, with one in three women reporting some form of violence (National Family Health Survey-5, 2021). Addressing intimate partner violence is central to global development agendas, including Sustainable Development Goal 5, which calls for the elimination of all forms of violence against women and girls. Rates are particularly high in rural areas where gender norms, economic dependency, and limited institutional support constrain women’s agency in marital conflict. While structural and cultural determinants of gender-based violence have been widely studied, far less is known about how everyday conflict resolution behaviors within intimate partnerships may reinforce or reduce psychological violence. This pilot study examined the association between women’s conflict resolution styles and psychological gender-based violence in rural Karnataka as preparatory groundwork for a larger community intervention. Forty-two married women participating in local self-help groups completed structured interviews assessing four conflict resolution styles – problem solving, engagement, compliance, and withdrawal – along with psychological violence, mood symptoms, financial stress, and husbands’ alcohol use. Consistent with socio-ecological and feminist theories, conflict styles reflecting low power and relational disengagement were most strongly linked to psychological violence. Women who reported complying during conflict (e.g., yielding quickly, suppressing their perspective) or withdrawing (e.g., remaining silent, disengaging) experienced higher levels of psychological aggression. Engagement behaviors (e.g., escalating or becoming angry) showed a weaker and nonsignificant association. In contrast, constructive problem solving was not related to lower psychological violence, suggesting that cooperative strategies may have limited protective value when embedded within broader structural inequalities and gendered power dynamics. Contextual vulnerabilities provided additional nuance. Financial worries and mood difficulties, although prevalent, were not significantly associated with psychological violence, indicating that emotional or economic strain alone may not predict risk in this small sample. Husbands’ alcohol use showed expected directional patterns: daily consumption was linked to higher psychological violence, while occasional use showed a modest inverse association. These findings align with prior research identifying alcohol as a catalyst for escalation and highlight its relevance for future intervention components. Reporting of physical violence was rare, underscoring ongoing challenges in disclosure and the need to revisit sampling strategies, interviewer training, and culturally sensitive framing before scaling up. Overall, the pilot study demonstrates that conflict disengagement – rather than overt aggression – is most strongly associated with women’s exposure to psychological violence. These results support an intervention model that integrates conflict-resolution skill building with community-level norm change, institutional responsiveness, and targeted modules on alcohol use and emotional regulation. Insights from this pilot directly inform the design, sequencing, and theoretical grounding of the forthcoming multi-level intervention aimed at reducing gender-based violence in rural Karnataka.

ABSTRACT Virtual faces are increasingly used as social partners in digital environments, yet it remains unclear whether the evaluative dimensions of ‘beauty’ and ‘attractiveness’ shape their perception in distinct ways. This eye tracking study examined how evaluation dimension (beauty vs. attractiveness) and viewing condition (3,000 ms vs. free viewing) influence ratings and gaze allocation for standardised Chinese virtual faces. Twenty four students rated 63 static faces for either beauty or attractiveness under both viewing conditions while their eye movements were recorded. Behaviourally, beauty and attractiveness judgments were almost indistinguishable, and mixed effects models showed no reliable effects of evaluation dimension. In contrast, gaze data revealed dimension dependent viewing strategies. Beauty judgments emphasised the core configural region, whereas attractiveness judgments shifted relatively more dwell time toward the mouth, indicating differences in information sampling rather than explicit evaluation. In free viewing, total face dwell time showed an inverted U shaped relation to ratings, cautioning against treating viewing duration as a proxy for preference. These findings indicate that, for standardised virtual faces, beauty and attractiveness converge at the level of explicit judgments but diverge in information sampling strategies, and indicate interpreting viewing time as a simple proxy for preference in virtual character research.

Diabetes mellitus remains one of the most prevalent chronic conditions worldwide, demanding continuous and accurate monitoring of blood glucose levels to prevent life-threatening complications such as hypoglycemia and hyperglycemia. Traditional glucose prediction systems rely solely on historical glucose readings, overlooking the physiological relationship between cardiovascular activity and glycemic fluctuations. In this work, a lightweight Sequential Transformer model is proposed that integrates heart rate signals alongside conventional diabetes management inputs — including basal insulin, bolus insulin, and carbohydrate intake — to achieve more physiologically informed blood glucose forecasting across multiple prediction horizons ranging from 5 to 30 minutes. The model is trained and evaluated on the OhioT1DM dataset comprising six Type-1 diabetic patients. To address the clinically critical problem of hypoglycemia detection, a Combined Focal-Asymmetric Huber Loss is introduced alongside a hypoglycemia-aware oversampling strategy. A post-training threshold calibration further tunes the decision boundary by maximising the F2-score on the validation set. The proposed system achieves a root mean square error of 5.81 mg/dL, a mean absolute percentage error of 1.28%, and an R² of 0.877, with 98.89% of predictions falling within the clinically safe Zone A of the Clarke Error Grid. Hypoglycemia sensitivity improved from 0% in the baseline to 66.7% at the critical 5-minute prediction horizon, demonstrating that targeted loss design and sampling strategies can transform a clinically unsafe model into a practically deployable glucose forecasting system.

Airfoil flow-field prediction is important for aerodynamic design, but wind-tunnel testing and computational fluid dynamics (CFD) remain costly and time-consuming. Deep learning enables fast inference, yet many existing models still rely on fixed grid representations, which may lead to insufficient learning in high-gradient regions and larger local errors. This study proposes Spatial Multilayer Perceptron (Spatial MLP) together with an Error-Gradient-Guided Data Sampling (EGDS) strategy for airfoil flow-field prediction. Spatial MLP adopts a coordinate-based point-wise prediction framework. A spatial decoder is introduced as an auxiliary branch to enhance global flow consistency during pretraining, while channel-wise multi-head attention is incorporated to improve cross-variable feature coupling. EGDS prioritizes physically informative points according to relative prediction error and gradient magnitude, while retaining random samples to preserve data diversity. Experiments on an independent test set show that Spatial MLP reduces the mean relative error (averaged over the velocity components u, v, and pressure p) by 15.2% relative to the MLP baseline. With EGDS, the overall mean relative error is further reduced by 34.5% relative to the MLP baseline. These results demonstrate that combining global consistency constraints with targeted sampling effectively improves both global prediction accuracy and local reconstruction quality in high-gradient flow regions.

Sparse-View CT (SVCT) reconstruction improves temporal resolution and reduces radiation dose, yet its clinical use is hindered by artifacts due to view reduction and domain shifts from scanner, protocol, or anatomical variations, leading to performance degradation in out-of-distribution (OOD) scenarios. We propose a Cross-Distribution Diffusion Priors-Driven Iterative Reconstruction (CDPIR) framework to tackle the OOD problem in SVCT. CDPIR integrates cross-distribution diffusion priors, derived from a Scalable Interpolant Transformer (SiT), with model-based iterative reconstruction methods. Specifically, we train a SiT backbone, an extension of the Diffusion Transformer (DiT) architecture, to establish a unified stochastic interpolant framework, leveraging Classifier-Free Guidance (CFG) across multiple datasets. By randomly dropping the conditioning with a null embedding during training, the model learns a more transferable cross-distribution prior that encourages domain-invariant anatomical structures while allowing domain-specific appearance modulation. During sampling, the globally sensitive transformer-based diffusion model exploits the cross-distribution prior within the unified stochastic interpolant framework, enabling flexible and stable control over multi-distribution-to-noise interpolation paths and decoupled sampling strategies, thereby improving adaptation to OOD reconstruction. By alternating between data fidelity and sampling updates, our model achieves state-of-the-art performance with superior detail preservation in SVCT reconstructions. Extensive experimental results demonstrate that CDPIR significantly outperforms existing approaches, particularly under OOD conditions, highlighting its robustness and potential clinical value in challenging imaging scenarios. The code is available at https://github.com/ Graeme-Lee/CDPIR.

The proliferation of short-form educational video platforms has facilitated the public's access to health information; however, no research has assessed the characteristics and quality of videos related to atrial septal defect. ASD was selected because it is one of the most common congenital heart defects encountered across the lifespan, often requiring repeated explanations regarding diagnosis, follow-up, timing of intervention, and long-term prognosis for patients and families. In addition, although short-form video studies have been conducted for several other diseases, ASD-related content on major Chinese short-video platforms has not been systematically evaluated. This study aimed to evaluate the quality and reliability of short videos related to atrial septal defect on TikTok and Bilibili. The Chinese term "atrial septal defect" was used to search for related videos on TikTok and Bilibili, and a predefined sampling strategy was used to screen the first 100 algorithm-ranked videos from each platform on October 21, 2025. This sample size was determined a priori to provide a feasible and standardized cross-sectional sample for manual content evaluation and to maintain comparability between platforms, rather than being based on a formal sample size calculation. Duplicate, irrelevant videos, and videos published within seven days were excluded to reduce the instability of early engagement indicators. As a result, 155 videos were included for analysis. The overall quality of these videos was assessed using the Global Quality Score (GQS), VIQI, PEMAT, JAMA Benchmark, and the DISCERN tool. These instruments were used to evaluate educational quality, reliability, transparency, understandability, and actionability, but they did not constitute a direct assessment of factual accuracy or potentially harmful medical content. Interobserver reliability was assessed for the independently retained duplicate GQS ratings using quadratic weighted kappa. Because social media search results are algorithm-ranked and may be affected by platform personalization, the included sample should be interpreted as a snapshot of highly visible ASD-related videos on the sampling day rather than an exhaustive representation of all available videos. Among the 155 videos (Bilibili: n = 70; TikTok: n = 85), significant differences were observed across all engagement and quality metrics. TikTok videos demonstrated significantly higher values for likes (median 415.0 [IQR 186.0 to 961.0] vs. 11 [IQR 5.0 to 52.75]), collections (131.0 [IQR 50.0 to 332.0] vs. 26.5 [IQR 8.0 to 106.75]), shares (124.0 [IQR 35.0 to 583.0] vs. 7.5 [IQR 2.0 to 29.0]), and comments (61.0 [IQR 17.0 to 492.0] vs. 1.0 [IQR 0.0 to 9.5]) (all P < 0.05). In contrast, the Bilibili group had longer video durations (in seconds) (639.0 [IQR 327.0 to 1,106.0] vs. 326.0 [IQR 216.9 to 429.0]; P < 0.001) and longer times since upload (in days) (356.0 [IQR 71.25 to 770.25] vs. 60.0 [IQR 37.0 to 102.0]; P < 0.001). Content quality assessments also differed, with TikTok videos having higher median DISCERN scores (23.788 vs. 22.786; P = 0.001), JAMA_Benchmark (1.918 vs. 1.714; P = 0.001) and GQS scores (3.635 vs. 3.314]; P = 0.006). However, the proportion of professional uploaders differed markedly between platforms (TikTok: 96.47% vs. Bilibili: 44.29%), and uploader-level analyses suggested that part of the observed quality advantage on TikTok may be attributable to uploader composition rather than platform characteristics alone. Importantly, this conclusion was based on the absolute position of the observed scores on their respective validated scales rather than on an arbitrary composite cutoff. Specifically, median GQS values of 3.314 and 3.635 on a 1-5 scale indicate moderate rather than high educational quality, whereas median JAMA Benchmark values of 1.714 and 1.918 on a 0-4 scale indicate that, on average, fewer than half of the transparency/reliability criteria were met. Because no universally accepted single "target quality score" exists across GQS, JAMA, DISCERN, PEMAT, and VIQI, each instrument was interpreted according to its own published scale direction and anchors. TikTok videos related to atrial septal defect are more engaging and of higher content quality than those on Bilibili. Overall, health-related videos on both platforms showed only moderate educational quality, with notable limitations in transparency, reliability, source attribution, and actionable guidance. Professionally produced content tended to perform better, although between-platform differences should be interpreted cautiously because of differences in uploader composition. These findings suggest that greater participation by health professionals may help improve the quality and reliability of online health information. As this study assessed informational quality rather than factual accuracy and was limited to Chinese-language videos from two platforms at a single time point, the results should be interpreted as a platform- and time-specific snapshot.

Large language models (LLMs) have been extensively applied in various recommendation scenarios, including bundle generation, thanks to their exceptional reasoning capabilities and comprehensive knowledge. However, exploiting large-scale LLMs for bundle generation introduces significant efficiency challenges—primarily high computational costs during fine-tuning and inference due to their massive parameterization. Knowledge distillation (KD) offers a promising solution by transferring expertise from large teacher models to more compact student models. This study systematically investigates KD approaches for bundle generation with the goal of minimizing computational demands while preserving performance. Specifically, we explore three critical research questions: (1) how does the format of distilled knowledge impact bundle generation performance? (2) to what extent does the quantity of distilled knowledge influence the performance? and (3) how do different ways of utilizing the distilled knowledge affect the performance? To support this investigation, we propose a comprehensive KD framework that (i) progressively extracts knowledge from raw data in increasingly complex forms, i.e., frequent patterns \(\rightarrow\) formalized rules \(\rightarrow\) deep thoughts; (ii) captures varying quantities of distilled knowledge through different sampling strategies, multi-domain accumulation, and multi-format aggregation; and (iii) exploits complementary LLM adaptation techniques—in-context learning, supervised fine-tuning and their combination—to leverage the distilled knowledge for domain-specific adaptation and enhanced efficiency in small student models. Through extensive experiments on multiple real-world datasets, we provide valuable insights into how knowledge format, quantity, and utilization methods collectively shape the performance of LLM-based bundle generation, which exhibits the significant potential of KD for more efficient yet effective LLM-based bundle generation.