Applied Intelligence最新文献

{"title":"MAST-LLM: A large language model-driven multimodal adaptive spatio-temporal framework for air traffic complexity prediction","authors":"Rui Cheng,&nbsp;Jianping Fan,&nbsp;Chao Zhang,&nbsp;Meiqin Wu,&nbsp;Ruixin Chen,&nbsp;Mingxuan Chai,&nbsp;Anna Wang","doi":"10.1007/s10489-026-07469-7","DOIUrl":"10.1007/s10489-026-07469-7","url":null,"abstract":"<div><p>Airspace complexity prediction remains challenging due to intertwined spatial, temporal, and modeling factors. Spatially, airspace sectors exhibit heterogeneous dependencies from both geographical adjacency and dynamic traffic flows. Temporally, complexity evolves through multi-scale dynamics with long-range dependencies. Modeling-wise, existing spatio-temporal frameworks rely on static graphs and shallow temporal alignment, limiting their capacity for joint spatio-temporal reasoning, particularly under high-complexity and long-horizon scenarios. To overcome these challenges, this study proposes MAST-LLM, a Multimodal Adaptive Spatio-Temporal framework enhanced by Large Language Models (LLMs). First, a Mixture-of-Experts (MoE)-based temporal alignment phase is introduced to capture heterogeneous temporal patterns and align domain-specific air traffic sequences with pre-trained LLM representations. Second, a full spatio-temporal fine-tuning phase is developed, integrating adaptive multimodal spatial representations, multi-scale temporal features, and a Bidirectional Spatio-Temporal Attention (BSTA) mechanism for joint spatial–temporal interaction modeling. Finally, by leveraging the global reasoning and long-range dependency modeling abilities of LLMs, MAST-LLM establishes a unified representation that captures both local dynamics and global contextual dependencies. Extensive experiments demonstrate that MAST-LLM achieves superior performance at short- and medium-term forecasting horizons and remains competitive at longer horizons, with improvements of up to 9.5 percentage points in accuracy and 12.9 percentage points in macro F1-score. Furthermore, comprehensive ablation, factor importance, sensitivity and visualization analyses verify the effectiveness and robustness of each core component.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034448","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Physics-informed Kolmogorov–Arnold network for fire source localization and temperature field reconstruction in multi-source tunnel fires","authors":"Yan Li,&nbsp;Bin Sun","doi":"10.1007/s10489-026-07504-7","DOIUrl":"10.1007/s10489-026-07504-7","url":null,"abstract":"<div><p>Tunnel fires with multiple fire sources pose significant challenges for sources localization and temperature field reconstruction. This study proposes a Multi-source Physics-informed Kolmogorov–Arnold Network (PIKAN) to solve this inverse problem using sparse thermocouple data. The proposed framework embeds the tunnel ceiling heat diffusion equation into a KAN-based neural network, in which radial basis function layers are incorporated to enhance nonlinear approximation capacity and ensure stable automatic differentiation. Specifically, the network comprises a temperature field network and multiple heat release rate networks, which are jointly trained to infer fire source locations, heat release characteristics, and the spatiotemporal temperature rise field. The method is validated using a full-scale double-source tunnel fire experiment and a three-source tunnel fire scenario simulated by Fire Dynamics Simulator. Compared with the conventional Multi-source PINN, PIKAN achieves higher localization accuracy and more reliable temperature reconstruction. In the double-source case, the localization errors are reduced to 0.89 m and 1.06 m, respectively. In the three-source case, the localization errors are controlled within 1.07 m, with the minimum error reaching 0.02 m. The reconstructed temperature fields agree well with experimental and numerical results. The results demonstrate that Multi-source PIKAN provides an accurate and physically consistent approach for multi-source fire inversion in tunnel fires.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149035262","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"BootDiffLights: Decentralized traffic signal control via diffusion-based reward generation","authors":"Binyu Hao,&nbsp;Jiajun He,&nbsp;Jiaoling Zheng,&nbsp;Wenan Dong,&nbsp;Qingyong Yang,&nbsp;Xiaojing Ma","doi":"10.1007/s10489-026-07487-5","DOIUrl":"10.1007/s10489-026-07487-5","url":null,"abstract":"<div><p>In multi-traffic signal control, the overarching objective is to optimize regional traffic metrics such as average queue length and travel time. Conventional approaches rely on credit assignment within multi-agent reinforcement learning (MARL) to distribute global rewards among individual signals. However, these methods face two fundamental limitations. First, they suffer from low sampling efficiency due to extensive trial-and-error interactions. Second, critic networks must possess strong extrapolation and generalization capabilities to interpret agent policies and allocate credit effectively. To address these challenges, we propose BootDiffLights, a novel framework that employs a conditional diffusion model to provide an implicit credit-assignment mechanism. By adopting a semi-offline MARL paradigm, BootDiffLights transforms the problem from credit assignment to black-box optimization. Consequently, the focus shifts from exploring optimal policies to generating optimal sample data. This shift enables the framework to improve both sampling efficiency and the optimization of regional traffic metrics. All experiments are conducted with 5 random seeds, and results are reported as mean ± standard deviation. Experimental evaluations on synthetic and real-world datasets demonstrate that BootDiffLights achieves superior performance compared to state-of-the-art baselines.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034706","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Iterative Demand Response Algorithm for Human-Computer Cooperation Integrating Knowledge of Large Language Model","authors":"Jun Yin","doi":"10.1007/s10489-026-07428-2","DOIUrl":"10.1007/s10489-026-07428-2","url":null,"abstract":"<div><p>When the requirements of human-computer interaction are complex, vague or involve specific background knowledge, it is impossible to accurately understand human needs or intentions. In order to achieve an efficient and accurate response to the requirements, an iterative requirement response algorithm of human-computer cooperation is proposed by combining the powerful knowledge representation and generation ability of large language model and the iterative optimization idea of human-computer cooperation. Establish an effective data collection mechanism, calculate the response results of user requirements, continuously optimize the algorithm according to the demand response structure, improve the intelligent response ability and user experience, collect the real-time response information, and get the results of collecting the iterative demand response information of human-computer cooperation. In the known finite domain, the corresponding feature statistical decision is obtained, and the load characteristics of non-equidistant nodes in man-machine cooperation network are extracted through the knowledge of large language model. According to the two-level attention matrix, the knowledge framework of large language model is constructed, and the iterative demand response strategy of man-machine cooperation in two stages is analyzed. The experimental results show that the algorithm can remove abnormal data when the threshold is 10, and the proportion of abnormal data in the three data sets reaches the lowest state. After applying this algorithm, the throughput and utilization rate have been significantly improved, and the response speed is faster.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034504","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 geometry problem solver based on heterogeneous graph","authors":"Haotian Zhang,&nbsp;Qi Jia,&nbsp;Lu Liu,&nbsp;Cong Xu,&nbsp;Liang Jin,&nbsp;Yuting Pan,&nbsp;Feiyu Chen,&nbsp;Yihua Wang,&nbsp;Yuhan Liu","doi":"10.1007/s10489-026-07471-z","DOIUrl":"10.1007/s10489-026-07471-z","url":null,"abstract":"<div><p>Geometry Problem Solving (GPS) is a challenging multimodal task that requires understanding both textual descriptions and geometric diagrams to derive mathematical solutions. However, existing methods struggle to model diagram structural information, align textual symbols with visual primitives, and handle the sparse distribution of geometric primitives in the diagram. To address these limitations, we propose a novel geometry problem solver that models geometric diagrams as heterogeneous graphs to explicitly capture the structural relationships among geometric primitives. We first introduce a graph construction method that represents each diagram as a heterogeneous graph. Building on this, we present the <b>Geo</b>metric <b>R</b>elation <b>H</b>eterogeneous <b>G</b>raph (Geo-RHG) dataset, which comprises geometric diagrams, corresponding textual descriptions, and their heterogeneous graph representations. To bridge the modality gap between text and diagrams, we employ a cross-modal pre-training strategy that explicitly aligns textual symbols with their corresponding visual primitives. We also develop a hierarchical fusion mechanism that incorporates structural cues into the visual features during the modality fusion stage. This integration enhances the model’s capacity to reason over geometric content and supports more accurate problem solving in multimodal contexts. Extensive experiments demonstrate that our approach achieves superior performance on GPS benchmarks, outperforming existing baselines in both accuracy and generalization. The dataset and code will be released at https://github.com/hzl7/GeoRHGraph upon publication.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034967","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Context-aware semantic similarity measurement for unsupervised word sense disambiguation","authors":"Jorge Martinez-Gil","doi":"10.1007/s10489-026-07492-8","DOIUrl":"10.1007/s10489-026-07492-8","url":null,"abstract":"<div><p>The issue of word sense ambiguity poses a significant challenge in natural language processing due to the scarcity of annotated data to feed machine learning models to face the challenge. Therefore, unsupervised word sense disambiguation methods have been developed to overcome that challenge without relying on annotated data. This research proposes a new context-aware approach to unsupervised word sense disambiguation, which provides a flexible mechanism for incorporating contextual information into the similarity measurement process. We experiment with a popular benchmark dataset to evaluate the proposed strategy and compare its performance with state-of-the-art unsupervised word sense disambiguation techniques. The experimental results indicate that our approach substantially enhances disambiguation accuracy and surpasses the performance of several existing techniques. Our findings underscore the significance of integrating contextual information in semantic similarity measurements to effectively manage word sense ambiguity in unsupervised scenarios. The source code of this approach is available at: https://github.com/jorge-martinez-gil/uwsd.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034992","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 clustering and ranking fusion for tourism recommendations: A data-driven approach to 5A scenic spot evaluation","authors":"Wen Li,&nbsp;Zeyu Xiao,&nbsp;Bin Yu,&nbsp;Zeshui Xu","doi":"10.1007/s10489-026-07488-4","DOIUrl":"10.1007/s10489-026-07488-4","url":null,"abstract":"<div><p>In the tourism sector, tourists often face high-dimensional evaluation data, which makes informed decision-making challenging. To improve the effectiveness of tourism recommendations and address the challenges faced by Multi-Attribute Decision-Making (MADM) methods in handling high-dimensional evaluation data, this study proposes a method that combines dynamic clustering with ranking fusion techniques. First, a consistency-driven dynamic clustering algorithm is employed to identify and group evaluation attributes with similar ranking patterns across scenic spots. Second, the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) is applied within each cluster to obtain local rankings of scenic spots based on the corresponding evaluation information. Finally, the Borda count method is used to fuse the local rankings across different clusters and obtain a global ranking. Through an empirical application to 5A scenic spots in China, the proposed method demonstrates a high degree of consistency with existing tourism ratings. This study provides an effective recommendation framework for tourism management, particularly in scenarios involving high-dimensional and heterogeneous evaluation data, and demonstrates its potential practical value.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034480","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"U-NetL2O: An explainable model for low-dose CT image reconstruction","authors":"Son Pham,&nbsp;Ngoc-Thao Nguyen,&nbsp;Bac Le","doi":"10.1007/s10489-026-07480-y","DOIUrl":"10.1007/s10489-026-07480-y","url":null,"abstract":"<div><p>This study presents an explainable optimization framework within the Learning-to-Optimize (L2O) paradigm for image reconstruction. The proposed method embeds a U-Net-based learned update operator into an unrolled Linearized ADMM scheme, enabling the integration of data-driven priors with explicit optimization constraints while preserving a structured iterative reconstruction process. Unlike conventional plug-and-play or end-to-end reconstruction approaches, the proposed framework maintains the optimization structure, improving interpretability while retaining the representational power of deep neural networks. To further enhance transparency, we introduce an optimization-aware explainable AI (XAI) approach based on heatmap visualizations that illustrate how structural information is progressively refined across optimization iterations. Additionally, empirical diagnostic indicators are integrated to monitor in-distribution consistency, providing statistical warning signals for heavily degraded reconstructions. Experimental results demonstrate competitive reconstruction performance while providing insights into the internal dynamics of the learned optimization process.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034481","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Two decades of machine learning in bioacoustics: a systematic review","authors":"Eduardo Mora-González,&nbsp;Antonio Canepa-Oneto,&nbsp;José Antonio Barbero-Aparicio,&nbsp;Álvar Arnaiz-González","doi":"10.1007/s10489-026-07454-0","DOIUrl":"10.1007/s10489-026-07454-0","url":null,"abstract":"<div><p>Computational bioacoustics has expanded rapidly over the past two decades, driven by advances in artificial intelligence that enable large-scale, automated analysis of animal vocalizations and soundscapes. This systematic review synthesizes 108 studies published between 2002 and 2023, revealing a marked increase in research activity since 2016 together with an expansion from predominantly bird-focused studies toward broader taxonomic coverage, including amphibians, insects, marine organisms, and multispecies soundscapes. The review develops a unified comparative framework that standardizes studies according to classification objectives, taxonomic groups, preprocessing strategies, learning paradigms, and algorithm families, enabling a comprehensive assessment of methodological trends across the literature. Supervised learning remains the dominant paradigm, while deep learning architectures, particularly Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), have become the most widely adopted approaches for bioacoustic classification. Traditional classifiers, including Support Vector Machines and Random Forests, continue to provide competitive performance in many applications, particularly when integrated into appropriate analytical workflows. The comparative analysis indicates that preprocessing strategies, including data augmentation, feature reduction, noise removal, and window segmentation, are consistently associated with high-performing classification pipelines. Mixed-effects analyses further suggest that reported classification performance is more strongly associated with specific algorithmic implementations than with broader algorithm families. Furthermore, significant interactions between methodological components indicate that classification performance cannot always be explained by individual factors considered in isolation. Overall, these findings indicate that successful bioacoustic classification depends on the coherent integration of preprocessing, feature representation, and learning algorithms within a complete analytical pipeline. Despite the substantial methodological progress achieved during the last decade, considerable heterogeneity persists in datasets, preprocessing procedures, evaluation protocols, and reported performance metrics, limiting reproducibility and direct comparison across studies. To address these challenges, we propose a set of good-practice recommendations emphasizing standardized dataset documentation, transparent methodological reporting, consistent evaluation frameworks, and clearly defined research objectives. Together, these recommendations aim to improve reproducibility, facilitate cross-study comparisons, and strengthen computational bioacoustics as a robust discipline supporting large-scale biodiversity monitoring and conservation.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034902","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"An interpretable spatio-temporal fusion operating conditions classification model for uncertain complex industrial processes","authors":"Bei Sun,&nbsp;Yao Wu,&nbsp;Peng Kong,&nbsp;Yijun Wang","doi":"10.1007/s10489-026-07443-3","DOIUrl":"10.1007/s10489-026-07443-3","url":null,"abstract":"<div><p>In complex industrial processes, precise and interpretable classification of operating conditions is crucial for improving production efficiency. However, inherent complexities, such as cause-effect temporal correlations from chemical reactions, material flows, and external environmental disturbances, lead to significant uncertainties and complex spatio-temporal coupling in classification. Furthermore, traditional data-driven methods often lack integration with process mechanisms, limiting their interpretability and validity in real-world applications. To address these challenges, this paper proposes a classification approach based on a Bayesian-optimized Self-Organizing Map model, incorporating spatio-temporal fusion within a comprehensive state-space framework. The framework is built by integrating inlet, reaction conditions and output states, thereby capturing a broad range of operating states represented in the collected data. Spatial feature extraction is performed using graph networks to capture dependencies among industrial variables. The extracted spatial features are then compensated into a temporal feature extraction network, which utilizes state space models and attention mechanisms to generate integrated spatio-temporal features. Subsequently, these integrated features are input into the Bayesian-optimized Self-Organizing Map model to improve classification reliability and accuracy. The clustering results are provided with physical interpretations through the integration of process mechanisms. Finally, a case study on an arsenic salt cobalt removal process demonstrates the effectiveness and interpretability of the proposed method under the evaluated conditions, while its practical deployment and broader applicability require further validation.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034903","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}