Applied Intelligence最新文献

As financial fraud becomes increasingly complex, effective detection methods are essential. Quantum Machine Learning (QML) introduces certain capabilities that may enhance both accuracy and efficiency in this area. This study examines how different quantum feature maps and ansatz configurations affect the performance of three QML-based classifiers, the Variational Quantum Classifier (VQC), the Sampler Quantum Neural Network (SQNN), and the Estimator Quantum Neural Network (EQNN), when applied to two non-normalized financial fraud datasets. Different quantum feature map and ansatz configurations are evaluated, revealing distinct performance patterns. The VQC consistently demonstrates strong classification results, achieving an F1-score of 0.88, while the SQNN also delivers promising outcomes. In contrast, the EQNN struggles to produce robust results, emphasizing the challenges presented by non-standardized data. Statistical validation using ANOVA confirms the significance of observed performance differences. Additionally, robustness tests on the best-performing models under five quantum noise types show that they maintain competitive performance, supporting their practical applicability. These findings highlight the importance of careful model configuration in QML-based financial fraud detection. By showing how specific feature maps and ansatz choices influence predictive success, this work guides researchers and practitioners in refining QML approaches for complex financial applications.

Hyper-relational knowledge graphs can enhance the intelligence, efficiency, and reliability of industrial production, facilitate equipment coordination, and optimize supply chains. However, due to current data and technology limitations, the construction of knowledge graphs in the industrial domain remains imperfect. Link prediction can effectively address this issue. Therefore, this paper constructs a hyper-relational knowledge graph for mine hoists to perform link prediction tasks. In traditional triple data sets, link prediction involves masking entities or relations to predict MRR and Hits@K. Although significant progress has been made in link prediction on traditional triple data sets, research on hyper-relational data sets is still lacking. This paper proposes a new method for solving link prediction problems in hyper-relational knowledge graphs—HyperKGLinker, which specifically addresses link prediction in hyper-relational knowledge graphs. This method innovatively integrates hyper-relational text data and hyper-relational graph data. Compared to baseline models, our model shows improved predictive performance across different data sets, indicating a significant enhancement in the accuracy and efficiency of link prediction in hyper-relational knowledge graphs. Future Research Work: (1) Integrate generative large language models like GPT to further optimize and expand the self-constructed hyper-relational knowledge graph of mine hoists, thereby exploring the generalization ability of HyperKGLinker. (2) Investigate methods for model interpretability to enable users to understand the model’s prediction results and decision processes, increasing its credibility and usability.

The increasing emphasis on traffic flow prediction in urban traffic management and planning is being driven by the continuous development of urbanization. This study aims to investigate the necessity of introducing graph structure information in data contexts with relatively simple road structures. We conduct experimental investigations to analyze the role of introducing graph structure information by comparing a graph neural network model with graph structure information and our proposed Spatial-Temporal Convolutional Long Short-Term Memory Network (ST-ConvLSTMNet) model, which is designed to capture complex spatiotemporal dependencies efficiently,​​ using six types of actual traffic data. Moreover, we analyze the application of various graph neural network models in traffic flow prediction, comparing their accuracies and computational efficiencies. Lastly, we provide a detailed explanation of the characteristics of data with relatively simple road structures and conduct further exploration of each module’s functionality through ablation studies. Codes are available at https://github.com/CC10969Peng.

Personality provides valuable insights into users’ emotions and behaviors, with applications in psychological counseling, personalized recommendations, and social analysis. However, most existing personality detection methods treat traits as independent variables and rely on simple aggregation, lacking psychological interpretability. To address this gap, we propose a Type Dynamics-driven Personality Detection Model (TPD), which integrates psychological theory with deep learning to capture the synergistic interactions among personality traits under social contexts. Specifically, TPD constructs a multi-level graph attention mechanism to jointly model semantic relations between user posts and interdependencies among traits, revealing the internal cognitive dynamics of personality expression. To further capture the influence of external environments on trait expression, TPD introduces a GRU-based attention mechanism that models how external stimuli shape trait manifestation over time. In addition, by identifying core posts representative of each trait and leveraging large language models (LLMs) for contextual explanation, TPD provides interpretable user-centric analyses of online environments. Extensive experiments conducted on two real-world datasets demonstrate superior performance over seven state-of-the-art methods for personality detection, establishing a psychologically grounded framework for social-media-based personality detection. Our code is available at https://github.com/lambdarw/TPD.

Outlier detection for manifold-valued functional data has garnered increasing research interest in machine learning communities due to its broad applications. However, the inherent nonlinear structures of manifolds render conventional linear operators ineffective, posing significant challenges for outlier detection on Riemannian manifolds. This paper develops a novel outlier detection framework for Riemannian manifold-valued functional data. First, we identify a clean subset of observations by estimating a robust Fréchet mean function and implementing a max-min clean subset selection procedure. Subsequently, we construct an outlyingness measure on the tangent space of the estimated robust Fréchet mean function, followed by the development of a new multiple hypothesis testing procedure that incorporates false discovery rate control. Numerical simulations demonstrate that our method exhibits superior accuracy in outlier detection compared to existing approaches that overlook the underlying nonlinear manifold structure. An application to real-world long-haul flight trajectories, which are modeled as curves on a two-dimensional sphere embedded in a three-dimensional Euclidean space, is also provided to illustrate the efficiency. By accounting for the Earth’s surface curvature instead of approximating trajectories as straight lines in flat space, our framework provides a more geometrically faithful solution for outlier detection in such manifold-valued functional data.

With ubiquitous growth and evolution of data in real life applications, processing and analyzing dynamic data has significant potential to drive problem-solving across various domains. Incremental three-way approximations methods have garnered increasing concerns due to their ability to effectively enhance knowledge maintenance efficiency in dynamic data environments. In this study, we propose incremental updating methods for three-way approximations within incomplete neighborhood multi-granularity rough sets (INMGRSs) under the scenarios of addition or deletion of attributes. We devise a novel matrix representation for the three-way approximations in INMGRSs with the assistance of matrix operations. By analyzing the changes in relevant matrices resulting from attribute variations across multi-granularity, we develop and analyze the matrix-based mechanisms for updating these matrices, reducing the effort of unnecessary re-computation. In particular, we present two incremental algorithms in according with these mechanisms. Through a comprehensive comparative analysis, we conduct experimental verification to assess the performance of the proposed algorithms against existing incremental algorithms.

Artificial Intelligence (AI) has become increasingly important in critical domains such as medicine, where accurate and interpretable decision-making is essential. However, many high-performing AI models operate as “black boxes”, limiting transparency and making it difficult for clinicians to understand or verify predictions. To address this challenge, we present an eXplainable Artificial Intelligence (XAI) framework that integrates a fuzzy rule-based classifier with genetic algorithms and 2-tuple linguistic representations. The method incrementally generates general fuzzy rules, introduces fuzzy exception rules to capture atypical cases, and applies rule selection and parameter tuning to enhance both accuracy and interpretability. Experiments on nine medical datasets demonstrate that our approach achieves competitive or superior accuracy compared to state-of-the-art algorithms, while requiring fewer rules. These results show that the method not only improves predictive performance but also provides clear, human-readable explanations for each decision, thereby increasing trust and facilitating its application in medical practice.

The quality of the input pictures is the primary factor influencing palmprint recognition systems. However, inadequate lighting, auditory disturbances, and variations in contrast might render identification uncertain. We present a new method for enhancing palmprint images called Multiscale Contrast-Limited Enhancement for Adaptive Palmprints (M-CLEAP). With the use of multiscale processing and dynamic equalizing of the histograms and contrast limitations, this technique successfully reduces noise and enhances palmprint features. We introduce VisionInceptNet, an advanced model architecture that integrates the optimal features of Vision Transformer (ViT) and InceptionV3 network. VisionInceptNet exhibits the highest identification accuracy; however, extensive evaluation on prevalent palmprint datasets demonstrates that the proposed M-CLEAP significantly enhances image quality. VisionInceptNet and M-CLEAP collaborate to establish a robust and effective framework for palmprint recognition that surpasses all existing deep learning frameworks and conventional approaches. This collaboration enhances the scalability and reliability of palmprint-based authentication in practical applications such as forensic investigation, secure access management, and identity verification.

Labelled data scarcity remains a longstanding challenge in hyperspectral image analysis, primarily due to high spectral dimensionality and the laborious nature of manual annotation. Self-supervised contrastive learning (SSCL) recently emerged as a promising approach to address this challenge due to its ability to learn robust representations by distinguishing similar and dissimilar data samples guided by the inherent properties of data rather than by labels. Our study builds upon a previously established two-stage patch-level, multi-label classification method for hyperspectral imagery using contrastive learning and further examines its performance on single-label and multi-label classification tasks under scenarios of limited training data. The methodology unfolds in two stages. Initially, we train an encoder with a projection network in a contrastive learning approach. Next, we fine-tune the pre-trained encoder with a classifier. Our empirical results on four public hyperspectral datasets demonstrate consistent improvements over fully supervised methods, boosting overall accuracy by up to (4.4%) in multi-label and (7.14%) in single-label tasks (e.g. on the Pavia University dataset). Performance remains competitive even under a (50%) reduction in labelled training data. Our qualitative analysis confirms that the contrastive-based encoder can produce well-separated representations for different classes and identify location-based features, even though it was not explicitly trained on spatial cues. This suggests the method’s potential to uncover implicit spatial information. These findings highlight the value of self-supervised contrastive learning for hyperspectral image classification, offering a promising avenue for handling data scarcity while enhancing predictive accuracy. They also highlight the method’s resilience in real-world scenarios where abundant labelled examples are often unavailable.