Concurrency and Computation-Practice & Experience最新文献

The identification of personal credit risk constitutes a fundamental concern within the realm of financial risk management. As the credit industry experiences significant growth, the precise evaluation of borrowers' credit risk and the mitigation of credit default risk have emerged as critical priorities for financial institutions and researchers worldwide. To enhance the ability to identify defaulting customers, this paper proposes a credit risk identification algorithm based on Bagging Fast Correlation-Based Filter with Temporal Convolutional Network (BaggingFCBF-TCN). This algorithm initially incorporates the feature selection approach inherent in the Bagging strategy to identify and filter the characteristics associated with defaulting customers, which serves to mitigate the bias in feature selection outcomes that may favor the majority class. Subsequently, it employs an enhanced Temporal Convolutional Network (TCN) classifier for the purpose of credit risk assessment, thereby improving the ability to discern both long-term and short-term dependencies present in personal credit data. The test results show that: (1) The BaggingFCBF-TCN algorithm significantly enhances the model's ability to identify defaulting customers, achieving optimal overall identification performance. (2) The results of the combination effect analysis indicate that the personal credit risk identification model constructed using the BaggingFCBF-TCN combination algorithm outperforms other combination algorithms in both the original dataset and the dataset after class balancing treatment.

As a critical link in the steel metallurgy process, the automation level of ladle masonry directly influences the service life of the ladle and metallurgical quality. Considering that the ladle masonry manipulator is susceptible to environmental disturbance, sensor noise, system parameter variations, and other disturbances in complex environments, this paper proposes an active disturbance rejection control method based on a fixed-time observer. A multi-power fixed-time extended state observer is constructed to ensure that complex disturbances converge to the equilibrium point within a fixed time, enabling accurate estimation of total system disturbances within a fixed-time framework. Simulation results demonstrate that this method effectively overcomes the degradation of observation performance in the extended state observer caused by large initial observation errors, thereby enhancing the trajectory tracking accuracy and robustness of the manipulator.

Regret is a complex negative emotion often associated with feelings of remorse, self-blame, and disappointment regarding past actions or decisions. It plays a significant role in various business and decision-making contexts and also has an impact on the health of individuals. This work aims at the detection of regret-one of the most important emotion. Existing research on regret detection has been predominantly limited to English content. It is observed that people find it easier to communicate their feelings effectively in their native or code-mixed languages. However, there is no work focusing on the detection of regret from text written in these languages. To address this gap, this paper first presents a novel dataset in Hindi using posts/comments written in Hindi or Hindi Roman script from multiple sources, incorporating both manual and automated annotation techniques to enhance the quality and consistency of data labeling. Then, it proposes a multi-class regret detection framework to detect regret and classify its domain. The proposed framework HRegBERT-CNN integrates a fine-tuned BERT(regret) model for Hindi with CNN using N-gram word embeddings, enabling it to capture local contextual features and complex patterns in the text effectively. Experimental results show that the HRegBERT-CNN model outperforms state-of-the-art models on the Hindi regret dataset by at least 3% and 5% for regret detection and domain identification tasks, respectively, in terms of macro F1-score.

In general, opinion mining indicates the process of evaluating the opinions of people on several topics that are accessible in text form. It is an important aspect of natural language processing as it sets up the effective planning and decision-making for businesses and users. Opinion mining can be performed more effectively and conveniently by initially carrying out subjectivity recognition, which entails recognizing the text as objective or subjective. This research comprises various steps, like preprocessing, feature extraction, data augmentation and opinion mining. The complete procedure was implemented in the Spark framework that utilizes a master–slave framework. The preprocessing step is done with methods, such as stop-word removal, stemming, and lemmatization. Afterwards, feature extraction is done by extracting sentiWordNet features and statistical features that involve capitalized words, exclamation marks, and hashtags. Followed by the data augmentation, the opinion mining phase uses a HAN–HDLTex approach proposed by the combination of HAN and HDLTex architectures. The experimentation is done for the proposed HAN–HDLTex model that shows better accuracy with a rate of 0.949, sensitivity with a rate of 0.969, and specificity with a rate of 0.939.

An attributed network encodes richer information through node and edge attributes. Attributed network representation learning (ANRL) seeks to obtain low-dimensional node embeddings by jointly modeling structural topology and attribute semantics. Graph neural network (GNN)-based methods, which leverage recursive message passing, have become the mainstream approach in this area. However, existing reviews provide limited systematic categorization and comparative analysis. In this paper, we classify existing GNN-based attributed network embedding methods into six categories: graph convolution network (GCN)-based methods, heterogeneous graph neural network-based methods, graph autoencoder-based methods, bidirectional encoder representations from transformers (BERT)-based methods, hyper-graph neural network (HGNN)-based methods, and Bayesian graph neural network-based methods. We not only summarize a large number of attributed net-work embedding methods but also analyze and compare these methods. Additionally, we introduce some typical application scenarios in this field. Finally, we discuss the challenges and highlight several future research directions.

Banana farming plays a crucial role in supporting the livelihoods of people in equatorial and tropical regions. Not only does it support local economies, but it also contributes significantly to food security. However, banana crops are frequently threatened by fungal infections, such as Cordana, Pestalotiopsis, and Sigatoka, which largely affect yield and quality. Intelligent computational methods that identify diseases at an early stage and evaluate severity can greatly enhance the timeliness and effectiveness of plant health interventions. In this context, the current study proposes an innovative integrated platform for automatic detection and estimation of the severity of banana diseases. A hybrid sampling technique, SMOTE-ENN, is applied for the first time to address class disparity of specimens while removing noisy and potentially mislabeled datasets. Four different Convolutional Neural Network (CNN) architectures have been proposed, labeled CNN1 through CNN4, with varying layer depths and hyperparameters to extract distinctive features from diseased leaf images. Following classification, color thresholding has been used, which uses both HSV and Lab color spaces to measure disease-specific severity; lesion regions are accurately segmented. The empirical evaluation demonstrated that the 5-layer CNN2 architecture obtained the best classification accuracy of 96.87%, outperforming the other CNN variants in multiple parameters, including sensitivity, specificity, precision, and F1 score. Furthermore, the time and space complexities of the models have been analyzed and compared to modern baselines to assess computational efficiency. Following established agricultural norms, the severity percentage is then used to assign a severity grade of the disease and suggest suitable fungicides. The proposed CNN-based integrated framework facilitates timely interventions, reduces excessive pesticide use, and supports sustainable banana cultivation.

Scene understanding is a fundamental prerequisite for autonomous robotic navigation and decision-making, particularly in hazardous environments such as tunnel construction and cable laying. The complex and unstructured nature of tunnel scenes poses significant challenges to human operators, making accurate scene comprehension essential. In this work, we propose a lightweight semantic segmentation network, SAE-DeepLabV3+, tailored for tunnel environments. The model introduces Shuffle Attention following the four branches of the ASPP module to enhance global context aggregation and enable adaptive focus on critical features during multi-scale representation learning. To ensure efficiency, MobileNetV2 is employed as the backbone to significantly reduce model parameters and memory consumption. Furthermore, we incorporate an improved Efficient Channel and Spatial Attention (ECSA) module in the decoder to boost feature refinement with minimal computational overhead. Extensive experiments on a self-constructed tunnel dataset demonstrate the effectiveness of our approach. SAE-DeepLabV3+ achieves a mean Intersection over Union (mIoU) of 83.67% and precision of 92.33%, outperforming existing methods such as Fcn, LR-ASPP, PspNet, Unet, and SegFormer. Compared to the baseline DeepLabV3+, our model achieves improvements of +4.67% mIoU and +5.63% precision, while the model has only 5.81 M parameters. The number of FLOPS is reduced by 113.98 G. The model achieves an FPS of 109.88 in the testing environment and 52.8 FPS in the PyQt-based interface. These results highlight the proposed model's strong balance between accuracy and efficiency, offering a practical solution for safe and reliable tunnel scene understanding in real-world applications. This makes it a highly suitable core perceptual component for autonomous systems operating in complex and safety-critical tunnel environments.

With the rapid development of the Internet of Vehicles (IoV) and location-based services (LBS), the privacy and security of trajectory data have become a top priority. Disclosure of trajectory privacy may pose many risks to users. To solve this problem, this paper proposes a vehicle trajectory data protection scheme combining regional realizability and deep learning (RDPP). Firstly, a regional realizability processing is proposed, which divides and covers geographical areas according to road network density and then defines the trajectory generation restrictions. Secondly, this paper proposed a combined regional realizability of the trajectory data generation model (RRP-TrajGAN) that can combine the trajectory generation restrictions to generate trajectory data that is in line with the real situation. Finally, the proposed personalized privacy budget allocation method based on the clustering and density method (CD-DP) is used to cluster the generated trajectory data, and a reasonable privacy budget is allocated to the trajectory data according to the clustering density attribute. Compared with more advanced schemes, this paper's approach uniquely combines regional realizability processing with deep generative models and density-based privacy budget allocation, achieving a balance between privacy and utility without sacrificing real-world feasibility. The experimental results show that compared with other existing schemes, the proposed scheme's degree of privacy protection is improved by 11.88%–39.82%, while data availability can be well guaranteed. In addition, the time complexity of the proposed scheme is $��O�\left(�n�\log �n�\right)�$, which is better than the comparison scheme.

The banking industry is experiencing a significant transformation due to the growing demand for digital services. Traditional monolithic architectures are increasingly inadequate for managing the complexity and scalability requirements of modern banking systems. Consequently, there is a need to explore alternative architectures that can offer the necessary flexibility, maintainability, and performance. We first present a typical monolithic architecture of a banking system and describe its drawbacks. Next, we propose a microservice-based architecture for the banking system and discuss how this proposed architecture addresses the challenges and shortcomings of both monolithic and microservice architectures. Finally, we examine the positive effects of addressing these challenges and drawbacks on system quality attributes to demonstrate the high quality of the proposed architecture. This paper aims to design a banking architecture based on microservice principles and patterns that effectively address the challenges posed by traditional monolithic architectures. The proposed architecture also incorporates lessons learned from previous microservices implementations to mitigate the drawbacks associated with this approach. The qualitative assessment we will discuss demonstrates how the proposed architecture contributes to improvements in several areas, including future development, human resource management, service continuity, continuous delivery, and latency. By achieving these improvements, our proposed architecture significantly enhances the quality attributes of the system, ultimately facilitating high-quality service delivery.