Applied Intelligence最新文献

{"title":"A further understanding for four types of concepts in incomplete contexts based on their connections","authors":"Xue Tian,&nbsp;Ruisi Ren,&nbsp;Ling Wei,&nbsp;Yanhong She","doi":"10.1007/s10489-026-07311-0","DOIUrl":"10.1007/s10489-026-07311-0","url":null,"abstract":"<div><p>In formal concept analysis, incomplete data are represented as incomplete formal contexts. To convey the meanings of “jointly certainly possessing (possessed)” and “jointly possibly possessing (possessed)”, partially-known formal concepts are proposed. Partially-known formal concepts have three specific forms: SE-ISI concept, ISE-SI concept and ISE-ISI concept, reflecting uncertain information of an incomplete context in different aspects. Since partially-known formal concepts and formal concepts in the least and greatest completions are all defined by the upper and lower derivation operators, there must be some connections among them. Exploring these connections can further help us acquire knowledge of raw data. In this paper, we first study the connections between partially-known formal concepts and formal concepts in the least and greatest completions, and then present methods to construct partially-known concept lattices based on formal concept lattices. Secondly, we examine the connections among the three kinds of partially-known concepts from the viewpoints of single concept and the set of extents (intents), demonstrating methods for mutual conversion among them. Finally, based on these connections, algorithms are designed to acquire partially-known formal concepts from formal concepts, and experiments are conducted to validate the feasibility as well as the stability of the algorithms under different data missing mechanisms. When the incomplete contexts contain relatively few concepts, the algorithm exhibits superior performance.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133387","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":"NMPPI: multi-scale relationship structure with effective pre-training for protein-protein interactions","authors":"Yan Kang,&nbsp;Hu Yuan,&nbsp;Xinchao Wang,&nbsp;Shengzhe Yao","doi":"10.1007/s10489-026-07339-2","DOIUrl":"10.1007/s10489-026-07339-2","url":null,"abstract":"<div><p>Protein-protein interactions (PPIs) are crucial to various cellular functions, including signal transduction, cell cycle progression, and metabolism. Current works on PPIs prediction have begun to utilize various data information, which renders them insufficient to address two significant limitations: i) how to effectively extract features when external data is unavailable? ii) how to mine multi-scale features of the PPI networks with limited computational overhead? To address these challenges, we propose a novel lightweight framework, NMPPI, which incorporates sequential and structural features without relying on various external datasets. We innovatively introduce the unsupervised algorithm, Node2vec, to extract structural information only from node sequences within the PPI network. Furthermore, we first design a supervised learning model integrating a Graph Isomorphism Network with a Graph-Mamba module to capture multi-scale relational structural features. The Graph-Mamba module can capture inter-node dependencies within the PPI network with linear computational complexity, utilizing a State Space Model, thereby facilitating the extraction of long-range dependencies. Extensive experimental results demonstrate that NMPPI outperforms state-of-the-art methods on two real PPI datasets. Additionally, NMPPI demonstrates superior generalization performance in cross-dataset scenarios while maintaining high computational efficiency, requiring only 27% of the training time and 36% of the parameter size compared to advanced methods. The code of NMPPI will be publicly released upon publication at https://github.com/yh2966674794/NMPPI.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133272","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 two-stage framework integrating group relative policy optimization with Tabu search for dynamic rescheduling in UAV testing","authors":"Zhibin Mao,&nbsp;Yan Gao,&nbsp;Qian Pu,&nbsp;Minghui Wang,&nbsp;Haikuo Shen","doi":"10.1007/s10489-026-07449-x","DOIUrl":"10.1007/s10489-026-07449-x","url":null,"abstract":"<div><p>Equipment failures can disrupt unmanned aerial vehicle (UAV) batch testing and delay downstream validation tasks. This paper studies a dynamic rescheduling problem in which the estimated repair time (ERT) of a failed item becomes available at the disruption time. We first reformulate the recovery problem as a static resource-constrained multi-project scheduling problem (RCMPSP). We then propose GRPO-TS, a two-stage hybrid method that combines Group Relative Policy Optimization (GRPO) and Tabu Search (TS). In the first stage, a GRPO policy generates an initial feasible schedule. The policy network uses a mixture-of-experts module and historical feature fusion to capture resource competition and precedence relations. In the second stage, TS refines the initial schedule through local neighborhood moves. Experiments on extended RCMPSP benchmark instances show that GRPO-TS achieves the lowest normalized average makespan among the tested priority-rule and deep reinforcement learning baselines. It also improves over classical metaheuristics on medium and large instances. Its online rescheduling time ranges from 0.376 to 1.837 s. Under ERT perturbations of up to ±30% on mixed-scale instances, the makespan increase does not exceed 9.14%. These results support the use of GRPO-TS for time-critical UAV batch testing under equipment failures.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133392","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":"Enhancing protein–ligand pose prediction via dual-branch deep learning with docking-oriented multi-component loss and iterative refinement","authors":"Md Khorshed Alam,&nbsp;Julia Rahman,&nbsp;M A Hakim Newton,&nbsp;Mohammed Eunus Ali","doi":"10.1007/s10489-026-07475-9","DOIUrl":"10.1007/s10489-026-07475-9","url":null,"abstract":"<div><p>Protein-ligand pose prediction is a core task in structure-based drug discovery because it determines how a ligand fits within a protein pocket and directly affects downstream virtual screening and lead-optimization workflows. Recent graph neural network (GNN) methods have shown promise for protein-ligand pose prediction, while improving the accuracy of an initial docked pose remains an important task. In this work, we present KTransPose, a GNN-based protein-ligand pose-refinement framework that combines a dual-branch architecture, a docking-oriented multi-component loss, and iterative refinement. The dual-branch architecture combines a global pose transformation branch that predicts rotation and translation with a local coordinate correction branch that predicts atom-wise coordinate adjustments. The multi-component loss combines root mean squared distance (RMSD) when computed without rotational or translational alignment, mean squared error (MSE) after alignment by the Kabsch algorithm, intraligand distance regularization, and a protein-ligand clash penalty. We instantiate the framework with several GNN architectures and find that a TransformerConv-based model provides the strongest overall performance. Using the same preprocessing, initial docked poses, and evaluation procedure, KTransPose improves upon MedusaGraph on both PDBbind-2020 and CASF-2016 benchmark datasets, reducing the mean RMSD from 5.08 Å to 4.47 Å and from 4.81 Å to 4.24 Å respectively, corresponding to improvements of approximately 12% in both datasets.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10489-026-07475-9.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133328","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Fault recognition approach applied to a rotating machine supported by hydrodynamic bearings using explainable artificial intelligence","authors":"Daniel F. Goncalves,&nbsp;Aldemir A. Cavallini Jr.,&nbsp;Valder Steffen Jr.","doi":"10.1007/s10489-026-07491-9","DOIUrl":"10.1007/s10489-026-07491-9","url":null,"abstract":"<div><p>Rotating machines are widely employed in modern industry. The growing demand for efficiency and operational durability drives the development of intelligent rotors equipped with Artificial Intelligence (AI) methods that promote the extension of equipment lifespan through proactive maintenance actions and reducing costs associated with unscheduled downtime. Some common failures in rotating machines, such as unbalance, misalignment, and cracks, have similar symptoms, making it difficult to diagnose accurately. The absence of labeled historical data and explainable models that are understandable to end users also makes using AI difficult in the industry environment. This work presents a fault diagnosis technique based on vibration analysis and explainable AI models in this context. The proposed methodology achieved high diagnostic performance, with precision, recall, and F1-score exceeding 98.9% in numerical tests and demonstrating robust accuracy in experimental scenarios using a reduced set of 12 features. The methodology is demonstrated with interpretable applications that facilitate real-time fault diagnosis and proactive maintenance in rotating machinery, showcasing its potential for industrial implementation. Vibration responses are obtained through a representative finite element (FE) model of a horizontal rotor and its corresponding test rig. Techniques for feature extraction and selection are also employed. Ensemble clustering is performed for novelty detection, while traditional supervised approaches are used for classification purposes. The explainable AI tool is used to interpret the results, revealing that features such as the second harmonic of the rotational speed (A2X-S1) and skewness (R08-S1), both extracted from plane S1 in axial direction, play a critical role in distinguishing between similar fault patterns. The combination of robust performance and the possibility of interpretation from the output models demonstrates the potential of the proposed approach for fault recognition in rotating systems.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10489-026-07491-9.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133271","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Safe maximin multiple graph learning for multi-view clustering","authors":"Naiyao Liang,&nbsp;Zuyuan Yang,&nbsp;Dan Xiang,&nbsp;Mingyang Liu,&nbsp;Jianbin Xiong,&nbsp;Junjie Yang","doi":"10.1007/s10489-026-07307-w","DOIUrl":"10.1007/s10489-026-07307-w","url":null,"abstract":"<div><p>Graph-based multi-view clustering (MVC) has attracted considerable research attention due to its effectiveness in fusing multi-view information. Although many graph-based MVC methods have been proposed, the safety of this kind of method is rarely considered, limiting their application. In this paper, we propose a novel graph-based multi-view clustering method, named safe maximin multiple graph learning, to obtain the safety in graph learning. The proposed model introduces the maximizing performance gain strategy into the graph learning while performing the multiple graph fusion. Meanwhile, we develop a theorem to ensure the safety of the proposed model, such that it can obtain safe clustering result over the baseline clustering result. To solve the proposed maximin optimization problem, an effective algorithm is developed, together with a remark to analyze the safety of the proposed algorithm. Experimental results indicate that the proposed method can achieve safe clustering results, and obtain competitive multi-view clustering performance, compared to state-of-the-art approaches.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133088","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 risk budgeting via large language models: A context-aware framework for adaptive portfolio allocation","authors":"Soyeong Lim,&nbsp;Hyoungmin Ahn,&nbsp;Taekyoung Lee,&nbsp;Gayeon Kim,&nbsp;Sunghun Lim,&nbsp;Insu Choi","doi":"10.1007/s10489-026-07501-w","DOIUrl":"10.1007/s10489-026-07501-w","url":null,"abstract":"<div><p>This study introduces a dynamic risk budgeting framework that employs large language models (LLMs) to determine time-varying risk budgets through contextual reasoning over structured market summaries. Conventional risk parity strategies allocate portfolio risk equally across assets under static budget constraints, thereby failing to adapt to evolving market regimes. While Hidden Markov Models (HMMs) offer regime-dependent adjustments, their discrete state spaces and Markovian transition assumptions limit how finely they can represent continuously evolving market conditions. The proposed framework utilizes GPT-4.1 to infer market regimes and compute relative risk intensities from financial data encoded in Token-Oriented Object Notation (TOON). Empirical evaluation on a universe of 55 U.S. equities spanning eleven sectors from 2023 to 2024 indicates that the LLM-based approach achieves a Sharpe ratio of 2.150, representing a 46.4% improvement over risk parity (1.469) and outperforming HMM-based risk budgeting (1.597). The analysis indicates that prompt design materially affects performance: focused directional guidance (volatility asymmetry) is the most consistent driver of risk-adjusted performance across specifications, and in the primary configuration it is most effective when combined with input dimensionality reduction, consistent with attention concentration effects, whereas providing comprehensive feature sets without targeted guidance produces only marginal improvements over the baseline. However, several important caveats apply: the backtest period was characterized by favorable market conditions, and preliminary experiments with alternative LLM architectures yielded substantially inferior results, suggesting that findings may not generalize across models. These results document both the potential and the limitations of foundation models in financial decision-making, offering evidence relevant to researchers exploring LLM applications in quantitative finance.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133108","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":"scDEAN: a lightweight adaptive fusion model for clustering scRNA-seq data","authors":"Subhashis Chatterjee,&nbsp;Rohit Bose","doi":"10.1007/s10489-026-07502-9","DOIUrl":"10.1007/s10489-026-07502-9","url":null,"abstract":"<div><p>Clustering single-cell RNA sequencing (scRNA-seq) data is essential for uncovering cellular heterogeneity. However, the high dimensionality, sparsity, and noise inherent to these data pose major analytical challenges. Existing methods struggle to jointly preserve gene expression variability and cellular topology, often relying on computationally expensive attention mechanisms or complex correction procedures. To address these issues, we propose scDEAN, a lightweight deep clustering framework that integrates gene expression and cellular similarity information through an adaptive and interpretable fusion strategy. scDEAN adopts a dual-encoder architecture, combining a negative binomial-based autoencoder to model the count matrix with a graph autoencoder built on a UMAP-derived fuzzy graph to capture fine-grained cellular geometry. Instead of attention, the two representations are dynamically fused using a softmax-gated mechanism with minimal parameter overhead. In this framework, the clustering module is decoupled from the representation learning module to mitigate conflicting optimization objectives. Clustering is formulated as a differentiable centroid optimization task with explicitly learnable centroids, enabling the learning of fuzzy membership assignments without repeated K-means reinitialization. A distance-based regularizer is incorporated to promote compact intra-cluster structure and improved inter-cluster separation. Experiments on multiple public scRNA-seq datasets show that scDEAN achieves competitive performance compared with existing scRNA-seq clustering methods, with improvements observed on several datasets and supported by statistical analyses. Beyond clustering accuracy, biological validation using the top 50 saliency-derived genes from the HNSCC dataset reveals enrichment of functionally relevant pathways, while survival analysis on the TCGA-HNSC cohort supports their clinical relevance. Overall, scDEAN provides a robust framework for scRNA-seq clustering that preserves biologically meaningful structure.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133107","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":"Visualized anomaly detection on Bitcoin transaction based on the E-GTNet fusion model and graph database","authors":"Fan Wu,&nbsp;Wei Yin,&nbsp;Xiaoyan Hu,&nbsp;Xinxin Luo","doi":"10.1007/s10489-026-07433-5","DOIUrl":"10.1007/s10489-026-07433-5","url":null,"abstract":"<div><p>The cryptocurrency market has been thriving since its inception, but due to its decentralization and anonymity, cryptocurrencies have become a preferred tool for criminals engaging in financial crimes. Research on detecting abnormal transactions in cryptocurrencies aids in combating illegal and criminal activities related to cryptocurrencies and provides insights and methods for regulatory authorities to monitor suspicious transaction behavior. This study constructs a heterogeneous transaction graph based on Bitcoin transaction records, incorporating temporal evolution features. We propose a novel hybrid graph neural network framework, E-GTNet (Edge-enhanced Graph Transformer Network), which integrates edge-aware attention mechanisms, graph convolutional structures, and temporal sequence modeling. The framework is designed to fully exploit edge-level features, node-neighborhood attention, and the temporal dynamics of transactions. To further validate the accuracy and interpretability of the proposed anomaly detection framework, we use the Neo4j graph database to visualize and trace anomalous transaction graphs identified by the model. Results show that anomaly structures evolve from linear chains to more complex and covert patterns over time. Moreover, the predicted anomalous transaction graphs closely mirror the structural patterns and temporal dynamics of ground-truth anomalous nodes labeled anomalies, validating the model’s effectiveness in both structural and temporal pattern learning.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149035065","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":"MKMed: Multi-Knowledge alignment framework for medication recommendation","authors":"Haixu Ma,&nbsp;Guanyong Wu,&nbsp;Shi Mu,&nbsp;Chen Li,&nbsp;Shunpan Liang","doi":"10.1007/s10489-026-07337-4","DOIUrl":"10.1007/s10489-026-07337-4","url":null,"abstract":"<div><p>Medication recommendation plays a critical role in healthcare by providing effective treatments based on patient’s electronic health records (EHR). Prior studies have shown that incorporating more medication-related knowledge significantly improves the quality of medication representations. However, not all medications are associated with multiple types of knowledge simultaneously. For example, some may have only textual descriptions but lack structured or molecular data. This imbalance in knowledge availability hampers the performance of existing models—a limitation we refer to as the “bucket effect” in medication recommendation. We further quantify its severity through a comprehensive statistical analysis on the distribution of modality coverage across medications. To address this issue, we propose a novel framework named <b>M</b>ulti-<b>K</b>nowledge <b>Med</b>ication recommendation (MKMed), which introduces a cross-modal medication encoder capable of aligning heterogeneous modalities into a unified representation space. Specifically, we pre-train the encoder using contrastive learning on five complementary knowledge modalities (text, image, structure, chemical properties, and knowledge graph), and integrate the resulting representations with patient EHR data to generate personalized medication recommendations. By effectively integrating incomplete and unevenly distributed knowledge sources, MKMed explicitly mitigates the bucket effect, enabling robust recommendation performance even in sparse knowledge settings. Extensive experiments on MIMIC-III and MIMIC-IV demonstrate that MKMed consistently outperforms state-of-the-art baselines across multiple evaluation metrics. Specifically, on MIMIC-III, MKMed achieves improvements of 1.9% in Jaccard similarity and 1.3% in PRAUC compared to the best-performing baseline. These results highlight the effectiveness of cross-modal knowledge alignment for medication representation learning, and suggest its potential to support more reliable and safer clinical decision-making in real-world healthcare scenarios.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149035456","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}