Knowledge graphs offer a structured framework for organizing accident causation knowledge and supporting maritime safety decision-making. However, existing construction methods based on deep learning rely heavily on large-scale annotated corpora, resulting in high costs and limited scalability. This study proposes a novel knowledge graph construction method based on ontology-constrained prompt engineering for large language models, eliminating the need for model fine-tuning. A ship collision accident ontology comprising 10 entity types and 8 relation types is established to define semantic boundaries, and a two-stage extraction strategy is designed for entity-attribute identification and triplet construction. By embedding the ontology into structured prompts, the framework achieves high-precision conversion from unstructured accident reports to structured knowledge using few-shot prompting with lightweight human verification. Experiments on 318 official Chinese investigation reports demonstrate that entity recognition attains an F1 score of 94.7%, while most relation types exceed 96% F1. Ablation studies confirm that ontology constraints yield a 105.8% improvement over the baseline, primarily by anchoring extraction to the predefined ontology schema and improving coverage and standardization. The resulting knowledge graph reveals characteristic patterns including elevated accident frequency during early morning hours, human factor dominance, and a strong association between management deficiencies and severe accident outcomes. These findings provide a low-cost, scalable knowledge foundation for maritime accident causation analysis and safety governance.
Building on our previous work in [1], this paper establishes the local well-posedness of the Schrödinger–KdV system in H−3/16×H−3/4 for the resonant case. Combining this result with those in [1], we obtain local well-posedness for the parameter range max{−3/4,s1−3}≤s2≤min{4s1,s1+2}, and this range is sharp in the sense of the contraction mapping argument.
Plant extracts are commonly incorporated into the processing of agricultural products to exert preservative effects. However, the molecular mechanisms underlying their decay-inhibiting effects remain poorly understood. Identifying plant extracts with potent anti-spoilage activity is especially crucial for mung bean sprouts (Vigna radiata), which are highly susceptible to rapid decay after harvest. In this study, we comparatively evaluated the efficacy of five extracts (mango seed MS, loquat leaf LL, rosemary R, horsetail H, mate tea MT) on preserving sprouts quality. LL extract achieved optimal quality retention, whereas R accelerated spoilage; MS, H, and MT produced intermediate effects. Integrated transcriptomic and metabolomic analyses revealed that LL extract maintained quality by: (i) reinforcing membrane integrity through the activation of antioxidant system and the suppression of lipid degradation; (ii) enhancing biosynthesis of phenylpropanoids and alkaloids, particularly methyl 4-hydroxycinnamate, naringenin and securinine, which inhibited enzymatic browning and microbial proliferation; and (iii) promoting fructose accumulation while downregulating genes associated with cell wall degradation, thereby delaying tissue softening and water loss. In contrast, R extract depleted these protective metabolites, exacerbated oxidative damage and browning. Notably, securinine, an endogenous metabolite sustained by LL, suppressed the growth of spoilage microorganisms and delayed sprout deterioration. Our work provides insights into plant extract–mediated postharvest quality maintenance and highlights securinine as a promising candidate for developing effective strategies to suppress decay in highly perishable vegetables.
Amid an increasingly competitive environment, fostering innovation among employees has emerged as a central concern for both academia and the business community. Drawing upon the conservation of resources (COR) theory, this study develops a multi-level framework to examine the underlying mechanisms and boundary conditions of competitive psychological climate (CPC) on employees' innovative work behavior. By gathering data from 386 hotel employees and their leaders spanning 60 teams throughout four provinces along China's southeastern coast, the results indicated that CPC facilitated employees' innovative work behavior by increasing team cooperative orientation (i.e., resource-gain pathway); simultaneously, it inhibited innovative work behavior by eliciting workplace anxiety (i.e., resource-depletion pathway). Furthermore, psychological capital buffered the resource-depletion pathway, while error tolerance reinforced the resource-gain pathway. The results advance theoretical insight into CPC's dual influence mechanism and offer practical guidance for motivating employees to innovate in the highly competitive hospitality sector.
In various machine learning tasks, Transformer-based models have attracted significant attention and achieved notable success. To apply Transformers to Collaborative Filtering (CF), we designed a contrastive learning framework called Transformer Collaborative Filtering (TransCF), in which the Transformer serves as a backbone to extract user preferences from global collaborative signals. During training, instead of traditional user-item pairs, we utilize the entire set of items interacted with by each user to capture high-order item-to-item relationships. Additionally, we integrate a Siamese contrastive learning objective to ensure consistent and coherent user representations across different latent spaces, which significantly enhances performance in sparse and cold-start scenarios. Extensive experiments across diverse domains (including E-commerce, Movie, and News) demonstrate that TransCF achieves state-of-the-art results with superior prediction quality. Furthermore, we provide a scalability analysis confirming O(n) efficiency for large-scale systems and utilize attention-based visualizations to offer intuitive insights into user preferences.
