International Journal of Intelligent Systems最新文献

Machine learning–based classifiers have been used for text classification, such as sentiment analysis, news classification, and toxic comment classification. However, supervised machine learning models often require large amounts of labeled data for training, and manual annotation is both labor-intensive and requires domain-specific knowledge, leading to relatively high annotation costs. To address this issue, we propose an approach that integrates large language models (LLMs) into an active learning framework, achieving high cross-task text classification performance without the need for any manually labeled data. Furthermore, compared to directly applying GPT for classification tasks, our approach retains over 93% of its classification performance while requiring only approximately 6% of the computational time and monetary cost, effectively balancing performance and resource efficiency. These findings provide new insights into the efficient utilization of LLMs and active learning algorithms in text classification tasks, paving the way for their broader application.

It is nowadays quite challenging to manage and control the supply chain concerning transparency, traceability, and zero-trust security. Digital technology such as blockchain has shown promising features to ease the global supply chain for tracking, tracing, and authenticity. This study critically examines the potential of blockchain technology and smart contracts to manage global supply chain sustainability. It also analyzes the inherent opportunities, benefits, and common barriers to deploying blockchain in the supply chain. Moreover, an overview of blockchain technology and its application in the various industries’ supply chain management is illustrated in this study. Furthermore, an application demo related to blockchain in the supply chain is provided within the scope of this study with the view to demonstrating how various transactions in the supply chain are executed with higher authenticity. The study is concluded with several future research propositions and directions that may provide insight into overcoming current challenges and the adoption of blockchain for the supply chain.

In the last 10 years, there has been a rise in the number of Arabic texts, which necessitates a more profound understanding of algorithms to efficiently understand and classify Arabic texts in many applications, like sentiment analysis. This paper presents a comprehensive review of recent developments in Arabic text classification (ATC) and Arabic text representation (ATR). We analyze the effectiveness of various models and techniques. Our review finds that while deep learning models, particularly transformer-based architectures, are increasingly effective for ATC, challenges such as dialectal variations and insufficient labeled datasets remain key obstacles. However, developing suitable representation models and designing classification algorithms is still challenging for researchers, especially in Arabic. A basic introduction to ATC is provided in this survey, including preprocessing, representation, dimensionality reduction (DR), and classification with many evaluation metrics. In addition, the survey includes a qualitative and quantitative study of the ATC’s existing works. Finally, we conclude this work by exploring the limitations of the existing methods. We also mention the open challenges related to ATC, which help researchers identify new directions and challenges for ATC.

Intelligent systems have been widely used in various fields. They generate a large number of high-dimensional time series monitoring data in the process of operation, which often hide various potential abnormal conditions, which bring hidden dangers to the stable operation of the system. Existing anomaly detection methods mainly focus on the sequence characteristics of time series data, but often ignore the correlation between different variables of multivariate data, and the detection efficiency is low when facing high-dimensional time series data. To solve the above problems, we propose a deep anomaly detection method based on graph neural network, and combined with the big data computing framework Apache Flink, we construct a real-time anomaly detection system for large-scale high-dimensional time series data. Experimental results on SWaT and WADI show that our proposed method can accurately detect anomalies in multivariate time series data, and can perform low-latency real-time anomaly detection on high-dimensional industrial streaming data.

Consumers increasingly post online reviews concerning products or services on the social media platforms. Online reviews have become a reliable data source for extracting consumer preferences. Importance-performance analysis (IPA) is widely used in preference analysis, but it normally ignores the effects of the performance of competitors as well as the asymmetric between requirements and satisfaction. Therefore, this study extends the IPA model and proposes a preference analysis method that considers asymmetric impact and competitors based on group wisdom and influence mining from online reviews. To do so, after identifying service attributes from online reviews, preferences hidden in massive online reviews are quantified using linguistic distribution assessments. Then, the influence of reviewers is measured by introducing both the relationship influence and the information influence from different types of reviewers as group wisdom to determine the performance of service attributes. The concept of competitive dominance degree is defined as the degree of competitive advantage relative to that of competitors under realistic contexts. The preference analysis method of this study reflects group wisdom and competitive environments more realistically. Its applicability and effectiveness have been testified in the hospitality industry.

Recent advances in transformer architectures have significantly improved performance in time-series forecasting. Despite the excellent performance of attention mechanisms in global modeling, they often overlook local correlations between seasonal cycles. Drawing on the idea of trend-seasonality decomposition, we design a seasonal cycle-aware time-series forecasting model (SCA-Net). This model uses a dual-branch extraction architecture to decompose time series into seasonal and trend components, modeling them based on their intrinsic features, thereby improving prediction accuracy and model interpretability. We propose a method combining global modeling and local feature extraction within seasonal cycles to capture the global view and explore latent features. Specifically, we introduce a frequency-domain attention mechanism for global modeling and use multiscale dilated convolution to capture local correlations within each cycle, ensuring more comprehensive and accurate feature extraction. For simpler trend components, we apply a regression method and merge the output with the seasonal components via residual connections. To improve seasonal cycle identification, we design an adaptive decomposition method that extracts trend components layer by layer, enabling better decomposition and more useful information extraction. Extensive experiments on eight classic datasets show that SCA-Net achieves a performance improvement of 12.1% in multivariate forecasting and 15.6% in univariate forecasting compared to the baseline.

This study proposes a system that uses unsupervised learning to autonomously identify sensor data which suggest that a machine may soon fail. The system predicts three failure modes in the servo motor of an injection machine by learning multivariate data from heterogeneous sensors. The unsupervised learning model predicted failures with an average F1 score of 0.9958. A case study in an actual shop verified the system’s practical applicability. This shop is a factory that runs 27 injection machines of various tonnages. Results confirmed the ease of retraining the unsupervised learning model and demonstrated its portability. The use of an unsupervised learning model eliminates the difficulties and dependencies associated with data acquisition for supervised learning models. The case study indicated that the use of the proposed failure-prediction program can reduce maintenance costs by up to $US 140,000/y. It can be applied to various machines across different industries.

Breast cancer is the primary cause of death for women around the world, necessitating the development of highly accurate, interpreted, and technologically advanced predictive approaches to support early diagnosis and treatment. In this research, we introduce a deep learning (DL) model for predicting breast cancer using both public and private datasets. The model uses the internet of things (IoT) to improve data collection and real-time monitoring, and it also uses the SMOTE method to resolve issues of class imbalance. The proposed model combines an explainable AI approach with SHAP values to ensure model interpretability. To identify the best DL algorithm for this method, we assess and compare six different DL algorithms: temporal convolutional networks (TCNs), neural factorization machines (NFMs), long short–term memory (LSTM) networks, recurrent neural networks (RNNs), gated recurrent units (GRUs), and deep kernel learning (DKL). IoT devices allow for the continuous acquisition of patient data, which, when integrated with our predictive models, improve the capacity for early detection. Reliable cancer detection relies on our method’s enhanced predictive accuracy and sensitivity. Furthermore, we offer crucial transparency in clinical settings by using SHAP to give detailed explanations of model decisions. By employing thorough statistical analysis and cross-validation, we guarantee that our model is resilient and can be applied to various patient populations. The results show that our proposed IoT integrated method has the potential to improve prediction performance and boost confidence in AI-powered medical diagnostics by making them more accessible and easier to use. From a performance perspective, the proposed approach, which uses the TCN algorithm and SMOTE, achieved the best accuracy for BC prediction. With the public dataset, the experimental results were 99.44%, 100.0%, 99.01%, 98.75%, 99.37%, and 99.89% for accuracy, sensitivity, specificity, precision, F1-score, and AUC, respectively. The experimental results for accuracy, sensitivity, specificity, precision, F1-score, and AUC using the private dataset were 97.33%, 93.33%, 100%, 100%, 96.55%, and 99.48%, respectively. On the other hand, with the combined datasets, the TCN algorithm achieved 100% for all performance metrics.

Vanilla Transformers focus on semantic relevance between mid- to high-level features and are not good at extracting smoke features, as they overlook subtle changes in low-level features like color, transparency, and texture, which are essential for smoke recognition. To address this, we propose the cross contrast patch embedding (CCPE) module based on the Swin Transformer. This module leverages multiscale spatial contrast information in both vertical and horizontal directions to enhance the network’s discrimination of underlying details. By combining cross contrast with the transformer, we exploit the advantages of the transformer in the global receptive field and context modeling while compensating for its inability to capture very low-level details, resulting in a more powerful backbone network tailored for smoke recognition tasks. In addition, we introduce the separable negative sampling mechanism (SNSM) to address supervision signal confusion during training and release the SKLFS-WildFire test dataset, the largest real-world wildfire test set to date, for systematic evaluation. Extensive testing and evaluation on the benchmark dataset FIgLib and the SKLFS-WildFire test dataset show significant performance improvements of the proposed method over baseline detection models.