Expert Systems with Applications最新文献

{"title":"Automatic fault interpretation method embedded with clustering task in 3D-UNet3+","authors":"Chunxia Zhang ,&nbsp;Qing Zou ,&nbsp;Jiangshe Zhang ,&nbsp;Yongjun Wang ,&nbsp;Lu Huang ,&nbsp;Chunfeng Tao","doi":"10.1016/j.eswa.2025.127704","DOIUrl":"10.1016/j.eswa.2025.127704","url":null,"abstract":"<div><div>To enhance the efficiency of subsurface data analysis, it is crucial to conduct precise interpretation of faults. Current research mainly focuses on the binary segmentation of faults, which often fails to accurately capture the intricate relationships between different faults. Therefore, this paper further carries out instance segmentation on the basis of binary segmentation of faults, focusing on how to effectively segment fault probability volume. To fulfill the data diversity of deep learning, we have created a labeled fault dataset containing 200 training sets and 20 validation sets based on synthetic data. Afterwards, we develop a 3D-UNet3+ network that fully integrates full-scale information, combined with mean shift clustering technology, to achieve fault instance segmentation. To guarantee precise differentiation among different fault instances, we select discriminative loss as the loss function for training. Extensively tested on synthetic and field data, our algorithm can complete the prediction of new data within tens of seconds and demonstrates excellent segmentation performance. In comparison to prevalent methodologies, our method not only improves segmentation precision but also significantly reduces the number of parameters, offering an innovative and more efficacious resolution for automatic fault interpretation.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127704"},"PeriodicalIF":7.5,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143869643","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A fresh view on Least Quantile of Squares Regression based on new optimization approaches","authors":"Justo Puerto,&nbsp;Alberto Torrejon","doi":"10.1016/j.eswa.2025.127705","DOIUrl":"10.1016/j.eswa.2025.127705","url":null,"abstract":"<div><div>Regression analysis is an important instrument to determine the effect of the explanatory variables on response variables. When outliers and bias errors are present, the standard weighted least squares estimator may perform poorly. For this reason, many alternative robust techniques have been studied in literature. In these terms, the Least Squares Quantile (LQS), and in particular the Least Squares Median, are among the regression estimators that exhibit better robustness properties. However, the accurate computation of this estimators is computationally demanding, resulting in a difficult estimator to obtain. In this paper, new novel approaches to compute a global optimal solution for the LQS estimator based on single-level and bilevel optimization methods are proposed. An extensive computational study is provided to support the efficiency of the methods considered, and an ad hoc procedure to address the scalability of the problem to larger instances is proposed.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127705"},"PeriodicalIF":7.5,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143870009","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"MGTNSyn: Molecular structure-aware graph transformer network with relational attention for drug synergy prediction","authors":"Yunjiong Liu ,&nbsp;Peiliang Zhang ,&nbsp;Dongyang Li ,&nbsp;Chao Che ,&nbsp;Bo Jin","doi":"10.1016/j.eswa.2025.127699","DOIUrl":"10.1016/j.eswa.2025.127699","url":null,"abstract":"<div><div>Accurately predicting the synergistic effects of drug combinations is a significant challenge for modern personalized oncology treatments. Graph neural networks (GNNs) can capture rich structural information about drug molecules, supporting the prediction of cancer drug responses and accelerating the discovery of novel drug combinations. However, the existing GNN-based methods have problems such as over-smoothing and over-squashing, which limit the techniques’ ability to express the structural information of drug molecules. To this end, this study proposes a molecular structure-aware graph Transformer network with relational attention for predicting drug synergy (MGTNSyn). MGTNSyn utilizes a graph relational attention network to aggregate key local substructures and identify molecule functional groups. It also employs the molecular structure-aware graph Transformer network to detect mutagenic motifs in drugs from a global perspective. The information on drug structure obtained at local and global levels enables MGTNSyn to better understand the mechanism of drug therapy for cancer. Extensive experiments on two real-world datasets demonstrate that MGTNSyn outperforms other state-of-the-art methods and alleviates expression limitations. The novel drug combination prediction experiments on three cancer cell lines demonstrate the method’s ability to discover therapeutic drug combinations.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127699"},"PeriodicalIF":7.5,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143869516","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SMDRL: Self-supervised mobile device representation learning framework for recording source identification from unlabeled data","authors":"Chunyan Zeng ,&nbsp;Yuhao Zhao ,&nbsp;Zhifeng Wang","doi":"10.1016/j.eswa.2025.127635","DOIUrl":"10.1016/j.eswa.2025.127635","url":null,"abstract":"<div><div>In mobile recording device source identification, deep learning techniques have been pivotal for extracting deep features from audio signals. Traditional approaches, however, predominantly rely on fully labeled datasets for supervised training, neglecting the vast amounts of unlabeled data typically present in real-world scenarios. This limitation has led to a significant disparity between the performance of existing models and their practical applicability. To bridge this gap, we introduce the Self-supervised Mobile Device Representation Learning (SMDRL) framework. During the pre-training phase, SMDRL utilizes a substantial unlabeled audio dataset enhanced by three novel data augmentation techniques: interpolative noise mixing, time-frequency masking, and partitioned resampling. Employing contrastive learning, these methods facilitate the development of a universal encoder capable of effectively capturing device-specific features from raw audio. Further, we propose the Cross-scale Mobile Device Encoder (CMDEncoder), which integrates a Convolutional Neural Network (CNN) for local feature extraction with Long Short-Term Memory (LSTM) and Transformer-Encoder architectures to handle global-scale information. This encoder is further refined by numerical computations of standard deviations and mean values to enhance global feature interaction. In the fine-tuning phase, the framework employs the Light Weight Enhanced Channel Attention, Propagation in Time-Delay Neural Network (LWECAP-TDNN) classifier, achieving high-precision identification results. Our experimental outcomes demonstrate that the proposed method significantly elevates the accuracy of device source identification, achieving recognition rates of 95.24% and 89.87% on the CCNU Mobile Base and CCNU Mobile Large datasets, respectively. These results represent improvements ranging from 1.44% to 35.69% over baseline methods. To facilitate further research, the source code for this study is made publicly available at <span><span>https://github.com/CCNUZFW/SMDRL</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127635"},"PeriodicalIF":7.5,"publicationDate":"2025-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143859160","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"KLLMs4Rec: Knowledge graph-enhanced LLMs sentiment extraction for personalized recommendations","authors":"Yachao Cui ,&nbsp;Kaiguang Wang ,&nbsp;Hongli Yu ,&nbsp;Xiaoxu Guo ,&nbsp;Han Cao","doi":"10.1016/j.eswa.2025.127430","DOIUrl":"10.1016/j.eswa.2025.127430","url":null,"abstract":"<div><div>Recommendation algorithms typically leverage auxiliary information such as user reviews and knowledge graphs to enhance algorithm performance, thereby alleviating data sparsity and cold start issues. Recently, researchers have increasingly employed large language models, which boast powerful natural language understanding capabilities, to further improve recommendation systems. However, these models often suffer from hallucination problems. Moreover, integrating heterogeneous information, such as reviews and knowledge graphs, can introduce new noise, potentially impairing recommendation performance. Knowledge graphs, as tightly organized structured knowledge bases, can assist in addressing the hallucination problem and heterogeneous information fusion problem of LLMs. To effectively address the aforementioned issues, we propose the Knowledge Graph-Enhanced Large Language Model Sentiment Extraction for the Personalized Recommendation Model (KLLMs4Rec). It aims to solve the LLMs hallucination problem and the noise problem caused by the fusion of heterogeneous information in recommender systems, and provide users with more accurate, diverse and novel personalized recommendations. To address the hallucination problem when extracting user sentiments from reviews with LLMs, we designed a knowledge graph-enhanced prompt template. It is worth noting that this scheme also solves the noise issue of heterogeneous information fusion. Additionally, to further expand user preferences extracted from reviews, this paper proposes a new hierarchical sentiment attention graph convolutional network, which utilizes three sentiment weight schemes to propagate user personalized preferences on the knowledge graph. Extensive experiments on the Movielens-20 m, Amazon-book, and Yelp datasets demonstrate that our model surpasses current leading methods while effectively addressing the hallucination problem of LLMs and the noise problem of heterogeneous information fusion.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127430"},"PeriodicalIF":7.5,"publicationDate":"2025-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143851835","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Shadow detection and removal for remote sensing images via multi-feature adaptive optimization and geometry-aware illumination compensation","authors":"Zhizheng Zhang ,&nbsp;Rui Cao ,&nbsp;Hongting Sheng ,&nbsp;Mingqiang Guo ,&nbsp;Zhenfeng Shao ,&nbsp;Liang Wu","doi":"10.1016/j.eswa.2025.127769","DOIUrl":"10.1016/j.eswa.2025.127769","url":null,"abstract":"<div><div>Shadows in remote sensing images degrade quality and obscure ground details, posing challenges in their accurate detection and removal. The biggest challenge in shadow removal is accurately detecting the shadow while restoring normal illumination. Therefore, this paper proposes a novel approach combining multi-feature adaptive optimization and geometry-aware illumination compensation for shadow detection and removal. The method introduces a novel multi-feature adaptive optimization algorithm, which simulates dynamic interaction behavior of snakes to obtain optimal shadow thresholds from multi-feature channels, achieving precise shadow detection. Then, Sunlit regions homogeneous to shadows are identified through irregular block matching, utilizing direction-adaptive feature extraction. Finally, we deduct geometry-aware illumination compensation theoretically to effectively remove shadows and restore normal lighting. Additionally, at the shadow boundaries, a Manhattan-based dynamic compensation method is designed to ensure smooth boundary transitions and mitigate pixel oversaturation. Experimental results demonstrate the superior performance of the proposed method compared to state-of-the-art methods of shadow detection and removal in both qualitative and quantitative ways. Overall, the proposed method provides a promising solution to the challenging problem of shadow in remote-sensing images. The code will be available at <span><span>https://github.com/whuzzzz/MAOSD</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127769"},"PeriodicalIF":7.5,"publicationDate":"2025-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143859188","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"UNITI: Framework for multi-task learning across datasets to mitigate overfitting","authors":"Seunghyun Kim ,&nbsp;Yeongje Park ,&nbsp;Eui Chul Lee","doi":"10.1016/j.eswa.2025.127653","DOIUrl":"10.1016/j.eswa.2025.127653","url":null,"abstract":"<div><div>Multi-task learning (MTL) has emerged as a promising approach for improving generalization across related tasks by leveraging shared representations. However, existing MTL techniques primarily focus on <strong>intra-dataset learning</strong>, where tasks share a common dataset with multiple labels. This approach often fails to address real-world scenarios where tasks originate from <strong>heterogeneous datasets</strong>, leading to catastrophic forgetting and feature interference. To overcome these limitations, we propose <strong>UNITI (Unifying Neural with Inter-dataset for Task Integration)</strong>, a novel <strong>inter-dataset multi-task learning framework</strong> that enables a single model to effectively learn from multiple distinct datasets. UNITI consists of <strong>two key components</strong>: (1) <strong>sequential dataset training</strong>, which reduces interference by updating model parameters systematically across datasets, and (2) <strong>feature-level knowledge distillation (KD)</strong>, where a student model learns essential task-related features from dataset-specific teacher models. We validate UNITI using both <strong>CNN-based (ResNet50) and ViT-based (SHViT) architectures</strong> on facial recognition tasks (age estimation, emotion classification) and general object classification (Caltech-101). Experimental results show that <strong>UNITI achieves up to 5.17% improvement in accuracy</strong> over standard MTL methods and maintains comparable performance to single-task models while significantly reducing computational overhead. Notably, in emotion recognition, UNITI improved accuracy from <strong>60.67% (single-task) to 65.84%</strong>, demonstrating its ability to preserve task-specific features in an inter-dataset setting. Our findings suggest that UNITI is a <strong>scalable and efficient alternative</strong> to traditional MTL approaches, with applications in real-world AI systems where diverse datasets must be integrated into a single model.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127653"},"PeriodicalIF":7.5,"publicationDate":"2025-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143864812","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Dynamic Meta-Decoupler-inspired Single-Universal Domain Generalization for Intelligent Fault Diagnosis","authors":"Mengdi Xu,&nbsp;Yingjie Zhang,&nbsp;Biliang Lu,&nbsp;Zhaolin Liu,&nbsp;Qingshuai Sun","doi":"10.1016/j.eswa.2025.127528","DOIUrl":"10.1016/j.eswa.2025.127528","url":null,"abstract":"<div><div>Rotating machinery in industry operates under complex conditions, with monitoring data influenced by irregular load fluctuations. Traditional domain generalization methods address distribution shifts using data from multi-source domains. However, it is time-consuming and expensive to collect data that covers all operating conditions and fault types. To overcome these limitations, this paper considers a more realistic yet challenging scenario called Single-Universal Domain Generalization (Single-UDG). It utilizes only single-source domain data to address the difficulties of unknown target domain data and unknown class recognition. We propose a novel learning framework called Dynamic Meta-Decoupler by decoupling domain-dynamic parameters. By adding Meta-Perturb and Parameters-Perturb strategies, Dynamic Meta-Decoupler is enforced to learn more robust shared features. Additionally, to fully tackle the challenges posed by Single-UDG, we propose a novel training strategy called Meta Generative Adversarial Network (MetaGAN). By utilizing Meta-Perturb-enhanced instances, our model is enhanced to generalize to unknown target domains and reject unknown faults. Extensive experiments conducted on two machinery datasets demonstrate that our model effectively addresses Single-UDG fault diagnosis under unknown working conditions.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127528"},"PeriodicalIF":7.5,"publicationDate":"2025-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143851832","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Disentangled adaptive fusion transformer using adversarial perturbation for egocentric action anticipation","authors":"Min Hyuk Kim ,&nbsp;Jong Won Jung ,&nbsp;Eun-Gi Lee ,&nbsp;Seok Bong Yoo","doi":"10.1016/j.eswa.2025.127648","DOIUrl":"10.1016/j.eswa.2025.127648","url":null,"abstract":"<div><div>In recent years, egocentric action anticipation for wearable egocentric cameras has gained significant attention due to its ability to interpret objects and behaviors from a first-person perspective. However, the field faces challenges due to uncertainties arising from several sources: action-irrelevant information, semantically mixed representations of behaviors and objects, and the abrupt motion of the user. To address these challenges, we propose Ego-<span><math><msup><mrow><mi>A</mi></mrow><mrow><mn>4</mn></mrow></msup></math></span> to enhance the robustness and reliability of egocentric action anticipation systems. First, Ego-<span><math><msup><mrow><mi>A</mi></mrow><mrow><mn>4</mn></mrow></msup></math></span> selectively extracts action-relevant information to make efficient use of additional data beyond visual information. Second, Ego-<span><math><msup><mrow><mi>A</mi></mrow><mrow><mn>4</mn></mrow></msup></math></span> generates effective disentangled representations for verbs and nouns using learnable behavior and object queries. Finally, Ego-<span><math><msup><mrow><mi>A</mi></mrow><mrow><mn>4</mn></mrow></msup></math></span> enhances the continuity between the present and future using adversarial perturbation. The experimental results on the EPIC-Kitchens-100 and EGTEA Gaze＋ datasets demonstrate that Ego-<span><math><msup><mrow><mi>A</mi></mrow><mrow><mn>4</mn></mrow></msup></math></span> outperforms existing methods in terms of mean top-5 recall and top-1 accuracy, respectively.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127648"},"PeriodicalIF":7.5,"publicationDate":"2025-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143859166","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"TFS-Net: Temporal first simulation network for video saliency prediction","authors":"Longyi Li ,&nbsp;Liyan Dong ,&nbsp;Hao Zhang ,&nbsp;Jun Qin ,&nbsp;Zhengtai Zhang ,&nbsp;Minghui Sun","doi":"10.1016/j.eswa.2025.127652","DOIUrl":"10.1016/j.eswa.2025.127652","url":null,"abstract":"<div><div>Video saliency prediction (VSP) plays a critical role in modern video processing systems by optimizing computational resource allocation and enhancing overall system performance. However, existing VSP methods either lack effective temporal modeling or incur high computational costs, particularly struggling with the initialization of video sequences. This paper presents TFS-Net, a novel temporal-first simulation network for VSP that integrates both static and dynamic modeling via parallel-optimized self-attention mechanisms. Specifically, TFS-Net addresses the challenge of initial frame processing with the innovative F31 algorithm and improves multi-scale spatiotemporal feature integration through a Hierarchical Decoder with Multi-dimensional Attention (HDMA). Drawing inspiration from primate saccadic behavior, the F31 algorithm optimizes processing efficiency during both training and inference phases, demonstrating particular effectiveness in unmanned aerial vehicle (UAV) real-time applications. Extensive evaluations on public datasets demonstrate that TFS-Net achieves significant improvements over state-of-the-art methods, with gains of 14.6%, 12.0%, and 11.2% in AUC-J, CC, and SIM metrics, respectively. Further experiments on UAV video analysis validate the model’s robustness and practicality in real-world scenarios.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"282 ","pages":"Article 127652"},"PeriodicalIF":7.5,"publicationDate":"2025-04-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143869517","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}