Temporal knowledge graph reasoning aims to infer missing or future facts by modeling the temporal evolution of relational events. Existing methods typically use fixed time windows or simple temporal decay functions to aggregate historical information, where the time window’s start and end, as well as the weights for historical events, are often manually preset. Such designs struggle to capture the temporal uncertainty prevalent in real-world scenarios and overlook the heterogeneous contributions of historical events to the reasoning task, thereby limiting the model’s performance and generalization. To address these issues, we propose a fuzzy adaptive temporal knowledge graph reasoning framework, FATKG. Specifically, to achieve a flexible representation of temporal uncertainty, FATKG introduces fuzzy temporal membership functions that model event timestamps as continuous fuzzy temporal neighborhoods. Building on this, a relation-aware adaptive time window mechanism is designed to dynamically determine the valid historical range. Additionally, FATKG estimates historical-event contributions through a time-semantic weighting scheme that combines adaptive temporal membership with query-conditioned semantic relevance, allowing the model to identify historical facts that are both temporally valid and semantically informative.Experimental results demonstrate that FATKG significantly outperforms mainstream methods, particularly on the high-frequency GDELT dataset. This validates the effectiveness of FATKG in handling temporal uncertainty and the heterogeneity of historical events.
Initial trust in public-facing medical consultation AI is formed during early encounters, yet the safety judgments underlying this process remain insufficiently understood. Using a vignette-based survey of prospective users from the general public (N = 525) and partial least squares structural equation modeling , this study tested a dual-safety model in which perceived clinical safety (PCS) and perceived data security (PDS) were specified as two proximal safety judgments associated with initial trust (IT) and advice-following intention (INT). PCS and PDS were both positively associated with IT, and IT was positively associated with INT. Perceived explainability and key-evidence visibility were positively associated with PCS, whereas perceived system performance and stability was strongly associated with PDS and privacy invasion concern was negatively associated with PDS. Structural assurance showed an unexpected negative association with PCS, and scenario risk perception was not significantly associated with PCS or IT despite successful risk manipulation. The core PCS/PDS–IT pathways remained stable in supplementary robustness analyses. These findings clarify dual safety judgments as proximal correlates of initial trust and provide human-factors-oriented evidence for future work on safety communication, data-use explanation, and calibrated trust in public-facing medical AI.
Edible salt is highly prone to caking during storage due to moisture absorption and recrystallization. This study systematically evaluates the anti-caking performance and mechanisms of four common agents—potassium ferrocyanide (PF), ferric ammonium citrate (FAC), silicon dioxide (SiO2), and calcium silicate (CaSiO3)—using molecular simulation, dynamic vapor sorption, isothermal and humidity cycling experiments, and low-field NMR. Results show that PF and FAC significantly reduce moisture absorption, while SiO2 and CaSiO3 act as moisture scavengers. SiO2 at 5 g/kg nearly eliminated caking, demonstrating the highest effectiveness. LF-NMR revealed that anti-caking agents alter water distribution and mobility, suppressing the formation of liquid and solid bridges. This work provides mechanistic insights and practical guidance for selecting and optimizing anti-caking agents in edible salt.
Traffic object detection is a core component of intelligent transportation systems, but real-time detectors must handle scale variation, occlusion, dense traffic, and complex background clutter. YOLO-based detectors still suffer from limited multi-scale representation capacity, semantic degradation during feature reconstruction, and redundant computation in the detection head. To tackle these limitations, we design AMKF-YOLO, an adaptive multi-convolutional kernel fusion-based YOLO detector with a hybrid cross-scale head. The methodology is motivated by the observations that robust traffic perception requires coordinated improvements in receptive-field diversity, spatial-detail reconstruction, and localization-oriented prediction, rather than isolated single-stage modifications. In this methodology, the backbone first enriches scale-aware contextual features, the neck reconstructs spatially coherent representations, and the detection head converts the fused features into accurate localization-oriented predictions. Accordingly, three complementary components of AMKF-YOLO are detailed as: a cross-scale parallel multi-kernel fusion module that employs parallel depth-wise convolutional branches and adaptive branch weighting to capture short-, medium-, and long-range contextual cues; a content-aware progressive dynamic upsampling operator that injects local spatial context before offset prediction to preserve fine-grained details and reduce semantic fragmentation; and a lightweight hybrid cross-scale fusion attention head that strengthens spatial-channel feature interaction and allocates more representational capacity to localization while reducing redundant detection-head computation. Notably, ablation and visualization analyses indicate that the gains arise from the interplay between feature extraction, feature reconstruction, and detection-head design. Extensive experiments on KITTI, VisDrone2019, and BDD100K show that AMKF-YOLO improves detection accuracy compared to the YOLO11 baseline across urban and aerial traffic scenarios while maintaining lightweight, real-time inference.
Calcium-iron composite modified biochar (FeCaBC) exhibits excellent pollutant adsorption capacity, but its synergistic effects on cadmium (Cd)-contaminated soil remediation and greenhouse gas emissions remain unclear. This study aimed to clarify the impacts of FeCaBC on soil available Cd, crop Cd uptake, soil organic carbon (SOC), and soil CO₂ emissions in Cd-contaminated soils. Soil incubation, CO₂ emissions, and potting experiments were integrated to evaluate the performance of FeCaBC. Fourier-transform infrared spectroscopy (FTIR), X-ray photoelectron spectroscopy (XPS), and structural equation modeling (SEM) were used to elucidate the underlying mechanisms. FeCaBC application significantly improved soil nutrient availability, increasing available P (114 Soil physicochemical property and nutrient availability improved after adding FeCaBC. Complexation and ion exchange were the direct pathways for FeCaBC to immobilize Cd. The addition of FeCaBC reduced CO₂ emissions by regulating the soil functional groups. Cd remediation and CO₂ reduction were simultaneously achieved in FeCaBC treated soil.
