Expert Systems with Applications最新文献

{"title":"MACR-afford: Weakly supervised multimodal affordance grounding via multi-branch attention enhancement and CoT multi-stage reasoning","authors":"Peiliang Wu ,&nbsp;Fengtao Sun ,&nbsp;Yifan Liu ,&nbsp;Shanyi Zhang ,&nbsp;Shiyu Wang ,&nbsp;Xiaohu Zhou ,&nbsp;Wenbai Chen","doi":"10.1016/j.eswa.2025.129929","DOIUrl":"10.1016/j.eswa.2025.129929","url":null,"abstract":"<div><div>Multimodal affordance grounding plays a crucial role in enabling computer systems to understand and recognize the functions and potential uses of objects. Affordance grounding involves not only identifying the shape and appearance of objects, but also understanding how they interact with the environment and users. However, current research faces challenges in accurately localizing affordance regions and addressing previously unseen scenarios. To address these challenges, we propose a weakly supervised multimodal affordance grounding framework, termed MACR-Afford, which combines a Multi-Branch Attention Enhancement module and a multi-stage Chain-of-Thought reasoning module. Under the guidance of a large language model, MACR-Afford improves the ability of intelligent agents to recognize and utilize objects in complex environments. First, we introduce a Multi-Branch Attention Enhancement (MBAE) module to improve the complementarity among object features. By enhancing cross-branch attention and extracting complementary discriminative features, MBAE enables more accurate localization of affordance regions. Subsequently, we introduce a Chain-of-Thought multi-stage reasoning module to generates general affordance knowledge units, which are used to guide the model in localizing affordance-relevant regions. Comprehensive experiments demonstrate that MACR-Afford consistently achieves superior performance in both seen and unseen scenarios, surpassing state-of-the-art baselines across multiple evaluation metrics. The code is publicly available at: <span><span>https://github.com/HuiHui-Robot/MACR-Afford</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"299 ","pages":"Article 129929"},"PeriodicalIF":7.5,"publicationDate":"2025-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145271161","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":"Segmentation, fusion, and representation: A novel approach to multi-label classification for long texts","authors":"Xin Wang,&nbsp;Junfeng Xiao,&nbsp;Wang Zhang,&nbsp;Tao Deng,&nbsp;Qian Wang","doi":"10.1016/j.eswa.2025.129825","DOIUrl":"10.1016/j.eswa.2025.129825","url":null,"abstract":"<div><div>Multi-label text classification (MLTC) is a vital task in natural language processing (NLP), often requiring high-quality text representations generated by pre-trained language models (PLMs). However, the inherent input length constraints of PLMs limit their capacity to handle long texts effectively. To address this challenge, we propose an innovative framework for multi-label long text classification. Our approach incorporates a dynamic text segmentation algorithm that optimally partitions long texts, thereby mitigating the input length limitations of PLMs. Additionally, we enhance both text and label representations by integrating external knowledge, modeling label co-occurrence relationships, and employing attention mechanisms. Extensive experiments conducted on diverse MLTC datasets demonstrate the superior performance of our method and uncover intricate relationships between texts and their associated labels. The code is available at <span><span>https://github.com/Coder-Jeffrey/SKFRL</span><svg><path></path></svg></span></div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"298 ","pages":"Article 129825"},"PeriodicalIF":7.5,"publicationDate":"2025-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145267701","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":"DC-RRG: Diagnosis-centered cascaded radiology report generation","authors":"Zihao Lin ,&nbsp;Zinan Hong ,&nbsp;Zijian Zhou ,&nbsp;Miaojing Shi ,&nbsp;Jin-Gang Yu ,&nbsp;Jingping Yun ,&nbsp;Shuangping Huang","doi":"10.1016/j.eswa.2025.129884","DOIUrl":"10.1016/j.eswa.2025.129884","url":null,"abstract":"<div><div>Radiology report generation (RRG) aims to automatically generate clinical reports from chest Xray images, offering potential benefits in aiding diagnosis and reducing clinician workload. However, this task remains challenging due to the need for generating diagnostically accurate and comprehensive reports. Previous methods neglect explicit and fine-grained diagnosis guidance, which is critical for addressing these challenges. To this end, we introduce a novel Diagnosis Centered method for Radiology Report Generation (DC-RRG), explicitly incorporating diagnostic observations to guide report generation. Specifically, our method utilizes a cascaded inference framework, initially using a large language model to generate diagnoses, which subsequently guide the generation of comprehensive reports. We design multi-modal prompts (<em>i.e.</em>, textual instructions, visual features, and medical knowledge features) to extend the Multi-Modal Large Language Model (MLLM) to generate accurate and comprehensive diagnostic reports. Additionally, we also develop a progressive training strategy that aligns modalities in two stages: first using reports for coarse clinical priors, then using diagnoses for fine-grained details. Extensive experiments on MIMIC-CXR and IU-Xray datasets demonstrate that DC-RRG surpasses all previous methods and achieves new state-of-the-art results.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"299 ","pages":"Article 129884"},"PeriodicalIF":7.5,"publicationDate":"2025-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145334112","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 fault-tolerant task offloading framework via large-scale multi-objective evolutionary optimization and game-based decision mechanism","authors":"Tingting Dong ,&nbsp;Jinbu Wen ,&nbsp;Fei Xue ,&nbsp;Yuge Geng ,&nbsp;Xingjuan Cai","doi":"10.1016/j.eswa.2025.129910","DOIUrl":"10.1016/j.eswa.2025.129910","url":null,"abstract":"<div><div>Large-scale multi-access edge computing (MEC) optimization is challenging due to high-dimensional decision spaces, conflicting objectives, nonstationary conditions, and failure-prone infrastructure. This paper presents an adaptive Mahalanobis distance-based large-scale multi-objective evolutionary algorithm with knowledge transfer and a two-layer encoding (<strong>ALC-LSMOEA-KT</strong>). The task-offloading model optimizes latency, energy, load balance, and failure risk under communication and computation constraints. A two-layer sparse encoding separates variable activation from value search, and a phase-aware evolution with Mahalanobis-guided covariance adaptation exploits inter-variable correlations while preserving diversity. A Stackelberg-based fault-tolerant migration module reassigns disrupted tasks to sustain robustness. Experiments on scalable multi-objective optimization Problems(SMOP)/large-scale multi-objective optimization problem(LSMOP) benchmarks and a realistic MEC simulator with dynamic arrivals, bandwidth variation, and injected failures show consistent gains in inverted generational distance (IGD), solution diversity, and robustness. The results indicate a scalable and reliable approach to MEC optimization under high dimensionality and uncertainty.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"298 ","pages":"Article 129910"},"PeriodicalIF":7.5,"publicationDate":"2025-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145268274","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":"KAN-MoDTI: Drug target interaction prediction based on Kolmogorov-Arnold network and multimodal feature fusion","authors":"Hui Liu ,&nbsp;Haoxin Jia ,&nbsp;Wenze Li ,&nbsp;Wei Li ,&nbsp;Yuting Yuan","doi":"10.1016/j.eswa.2025.129828","DOIUrl":"10.1016/j.eswa.2025.129828","url":null,"abstract":"<div><div>Drug-target interaction (DTI) prediction is a crucial task in computational drug discovery and repurposing, as it accelerates candidate identification while reducing development costs. Despite the advancements in deep learning, existing methods still face challenges in effectively modeling multi-modal data, fusing heterogeneous features, and capturing complex nonlinear relationships. We propose KAN-MoDTI to tackle these challenges by integrating Kolmogorov-Arnold Networks (KAN) with multimodal feature fusion and adaptive gating mechanisms, effectively combining heterogeneous drug and target representations to better capture the complex interactions between them. In the feature encoding stage, we use a dual-branch approach: For drugs, we combine SMILES sequence embeddings with structural representations from a KAN-based graph encoder. For targets, we integrate N-gram sequence embeddings with biochemical descriptor features. In the feature fusion stage, we introduce the FeatureFusionKAN module, which uses a gating mechanism to assign adaptive weights and KAN to perform the integration of heterogeneous modal features. KAN is also utilized in the final prediction layer to enhance the model’s ability to accurately predict complex drug-target interactions. Comprehensive experiments on datasets such as DrugBank, BindingDB, and Human show that KAN-MoDTI consistently outperforms or matches recent state-of-the-art baselines across metrics like AUROC and AUPRC.The source code implementation can be found at: <span><span>https://github.com/jiahaoxin/KAN-MoDTI</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"298 ","pages":"Article 129828"},"PeriodicalIF":7.5,"publicationDate":"2025-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145268542","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 heterogeneous agent reinforcement learning approach with curriculum learning for variable speed limit control","authors":"Zhaoqing Li ,&nbsp;Silai Chen ,&nbsp;Guosheng Xiao ,&nbsp;Yangsheng Jiang ,&nbsp;Zhihong Yao ,&nbsp;Puxin Yang","doi":"10.1016/j.eswa.2025.129945","DOIUrl":"10.1016/j.eswa.2025.129945","url":null,"abstract":"<div><div>The majority of existing Variable Speed Limit (VSL) studies employ rule-based control strategies, which often fail to accommodate diverse road characteristics and lack responsiveness to sudden congestion. These limitations are further compounded by computational inefficiencies, hindering real-time decision-making in large-scale or complex traffic environments. To address these issues, this study proposes a VSL strategy based on Heterogeneous Agent Reinforcement Learning with Curriculum Learning (HARLCL). Specifically, a top-level agent classifies road segments using the Mini-Batch K-means algorithm, thereby capturing the heterogeneity of each segment. The lower-level agents operate with class-specific observation spaces and action spaces, each agent observes and acts its feature set tailored to its segment class while sharing the same reward design and learning architecture. Subsequently, lower-level agents adaptively adjust both the position and length of each controlled segment according to classification results and real-time traffic conditions. Through a reward-driven mechanism, these agents continually refine their control precision. Moreover, curriculum learning is introduced during the multi-agent training process, effectively accelerating convergence and mitigating computational burdens typically encountered in large-scale reinforcement learning. Experiments were conducted in three task scenarios–typical fixed bottlenecks, random task bottlenecks, and multiple bottlenecks using realistic road data. The results indicate that training time decreased by 43.18 % and 47.35 %; in the SUMO microscopic simulation, total travel time was 312.25 veh·h, 43.05 % lower than NoVSL, and safe braking distances also decreased, indicating improved safety and more stable traffic flow. Compared with conventional feedback control and deep reinforcement learning approaches, the HARLCL-based strategy demonstrates substantial advantages in training efficiency and control precision, offering a promising avenue for practical VSL implementation.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"299 ","pages":"Article 129945"},"PeriodicalIF":7.5,"publicationDate":"2025-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145334065","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":"AirMamba: A deep learning framework for long-term PM2.5 forecasting integrating multi-scale correlations and time-frequency dynamics","authors":"Jie Lian ,&nbsp;Xiao Wang ,&nbsp;Sirong Huang ,&nbsp;Dong Wang ,&nbsp;Qin Zhao","doi":"10.1016/j.eswa.2025.129937","DOIUrl":"10.1016/j.eswa.2025.129937","url":null,"abstract":"<div><div>Existing approaches for long-term forecasting of <span><math><msub><mtext>PM</mtext><mrow><mn>2.5</mn></mrow></msub></math></span> typically focus either on time-domain or frequency-domain features in isolation, neglecting their complementary interactions. This limitation restricts their capacity to effectively capture long-term trends. Moreover, the absence of explicit modeling of multi-scale correlations among influencing factors under complex environmental conditions may undermine both the stability and accuracy of model predictions. To overcome these limitations, we introduce AirMamba, a novel deep learning framework designed to enhance long-term <span><math><msub><mtext>PM</mtext><mrow><mn>2.5</mn></mrow></msub></math></span> forecasting by integrating multi-scale correlation analysis with time-frequency interactions. Specifically, a multi-scale inter-variable correlations extractor module is developed to capture the complex interdependencies among variables across diverse temporal scales. The framework leverages the Maximum Overlap Discrete Wavelet Transform (MODWT) to decompose time series data into multi-scale high-frequency and low-frequency components, thereby facilitating a comprehensive time-frequency analysis. An enhanced bidirectional Mamba structure is then employed to model both long- and short-term dependencies within the time series, informed by the identified time-frequency interactions. Extensive experiments demonstrate that the proposed method achieves superior forecasting performance compared to existing mainstream models.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"299 ","pages":"Article 129937"},"PeriodicalIF":7.5,"publicationDate":"2025-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145271450","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 hybrid deep learning rainfall-runoff forecasting model Incorporating spatiotemporal information from multi-source data","authors":"Wan Liu ,&nbsp;Li Mo ,&nbsp;Xiaodong Li ,&nbsp;Wenjing Xiao ,&nbsp;Haodong Huang ,&nbsp;Yongchuan Zhang","doi":"10.1016/j.eswa.2025.129974","DOIUrl":"10.1016/j.eswa.2025.129974","url":null,"abstract":"<div><div>Deep learning has been widely applied in runoff forecasting, focusing primarily on temporal features but neglecting the influence of spatial heterogeneity. Capturing complex spatiotemporal atmosphere–land–hydrology interactions by deep learning remains challenging in rainfall–runoff forecasting. This study proposes a hybrid deep learning framework, Runoff Forecasting Model Integrating Spatiotemporal Features (RFMISF), which leverages the complementary strengths of multiple deep learning architectures to construct five modules, thereby fusing multi-source data. Specifically, the framework integrates the Convolutional Neural Networks for extracting spatial features of underlying surface, the LSTM for capturing temporal dependencies in rainfall and runoff, and the Convolutional LSTM (ConvLSTM) for learning spatiotemporal features of meteorological inputs. Two case studies of daily runoff forecasting have been deviced for the BHT and SBY hydrological stations with distinct hydrological regimes. At the BHT, the RFMISF reduced RMSE by 31.53% and MAE by 33.39% compared to the Xinanjiang baseline; at the SBY, the RFMISF improved NSE by 13.6% and decreased MAE by 27.39%. Ablation experiments of excluding station rainfall, underlying surface, and meteorological data are further conducted to underline the importance of multi-source data. At the BHT, the experiments led to RMSE increases of 9.29%, 4.69%, and 5.59% during flood season, respectively. At the SBY, the experiments resulted in reductions of NSE by 15.08%, 4.46%, and 12.94%. Additionally, model performance varies with rainfall intensity, indicating the differentiated contributions of multi-source data in complex runoff responses. Although reanalysis data enhance spatial representativeness, their systematic errors require careful treatment. Overall, this study introduces a novel, robust framework for enhancing runoff prediction and improving water resource management in hydrologically complex environments.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"298 ","pages":"Article 129974"},"PeriodicalIF":7.5,"publicationDate":"2025-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145268421","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":"INN-RAE: Reversible adversarial examples based on invertible neural networks for facial protection","authors":"Zeyu Zhao,&nbsp;Ke Xu,&nbsp;Laijin Meng,&nbsp;Tanfeng Sun,&nbsp;Xinghao Jiang","doi":"10.1016/j.eswa.2025.129962","DOIUrl":"10.1016/j.eswa.2025.129962","url":null,"abstract":"<div><div>Reversible adversarial examples can effectively prevent data from being accessed and recognized by unauthorized deep neural network models, but existing methods struggle to balance the visual quality and attack effectiveness of the generated adversarial examples. This paper proposes a method for generating reversible adversarial examples based on invertible neural networks (INN-RAE), achieving an effective unification of high attack success rate and high visual stealthiness. Specifically, during the forward propagation phase of the invertible neural network, both the clean sample and a noise matrix are input simultaneously, and adversarial examples are generated by fine-tuning the noise matrix. When restoring the adversarial examples, the same invertible neural network can be used to achieve high-quality restoration and remove the attack noise, thereby realizing end-to-end reversible adversarial example generation and restoration. Compared with existing reversible adversarial example generation algorithms, INN-RAE achieves state-of-the-art levels of attack success rate on multiple face datasets and face recognition models, while also achieving better visual stealthiness and restoration effects.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"299 ","pages":"Article 129962"},"PeriodicalIF":7.5,"publicationDate":"2025-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145334058","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":"StructCare: Dynamic graph structure learning for enhancing context-aware healthcare","authors":"Qizheng Sun ,&nbsp;Xiang Li ,&nbsp;Chuankun Duan ,&nbsp;Chen Li ,&nbsp;Shunpan Liang","doi":"10.1016/j.eswa.2025.129942","DOIUrl":"10.1016/j.eswa.2025.129942","url":null,"abstract":"<div><div>The advancement of intelligent healthcare systems has generated large-scale electronic health records (EHRs). This data provides a foundation for advancing machine learning in healthcare, particularly for pivotal tasks like disease prediction and medication recommendation. Despite the growing focus on modeling relationships within EHR, most existing methods treat medical events as connected by static relationships, where associations are defined by co-occurrence or fixed correlations. However, such approaches overlook dynamic relationships that reflect how a patient’s health status and treatment evolve over time. Therefore, we propose StructCare, a clustering-based graph structure learning framework that explicitly models these dynamic relationships by integrating lab test results with medical events. Specifically, StructCare leverages large language models to construct relationships among medical events and generate personalized patient graphs from visit records. It then employs temporal networks to capture evolving health trends from monitoring-level events and dynamically updates the graph structure at each visit, generating richer patient representations, thereby improving the accuracy of disease prediction and medication recommendation. Extensive experiments on two real-world datasets and across two tasks show that StructCare consistently outperforms state-of-the-art methods. Specifically, it achieves an average improvement of approximately 2.1 % in F1-score and 2.8 % in Jaccard for disease prediction, and 2.3 % in F1-score and 4.0 % in Jaccard for medication recommendation, highlighting its effectiveness in capturing dynamic, context-aware relationships.</div></div>","PeriodicalId":50461,"journal":{"name":"Expert Systems with Applications","volume":"299 ","pages":"Article 129942"},"PeriodicalIF":7.5,"publicationDate":"2025-10-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145334169","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}