Knowledge-Based Systems最新文献

{"title":"IBL-AE: An interpretable base learning autoencoder for intelligent fault diagnosis of rotating machinery","authors":"Hongkun Li ,&nbsp;Chen Yang ,&nbsp;Bo Han ,&nbsp;Xiaoyu Cao","doi":"10.1016/j.knosys.2025.114687","DOIUrl":"10.1016/j.knosys.2025.114687","url":null,"abstract":"<div><div>Deep learning has demonstrated powerful capabilities in fault diagnosis, yet the opaque nature of its internal decision-making severely limits its application in critical engineering scenarios. To address this issue, we propose IBL-AE (Interpretable Basis Learning Autoencoder), a novel deep learning architecture that integrates non-negative basis learning into an autoencoder framework for explainable fault classification. IBL-AE incorporates a non-negative decomposition module inspired by non-negative matrix factorization (NMF) within the latent space, enabling the learned features to be explicitly associated with physically interpretable basis components. Unlike post-hoc interpretability techniques, IBL-AE achieves inherent interpretability by design, as both the network weights and outputs can be visualized and directly linked to key frequency bands indicative of specific fault types. A classification module further utilizes the learned coefficients to make decisions in a human-understandable manner. Extensive experiments on three rotating machinery datasets demonstrate that IBL-AE not only achieves diagnostic accuracy, but also offers interpretable and physically meaningful insights into model behavior, paving the way for more trustworthy and practical deployment in industrial fault diagnosis.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114687"},"PeriodicalIF":7.6,"publicationDate":"2025-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145363092","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":"BGICR: Bootstrap-guided iterative clustering refinement for enhanced high-dimensional psychological data analysis","authors":"Khoula Al. Abri,&nbsp;Manjit Singh Sidhu,&nbsp;Faridah Hani Mohamed Salleh","doi":"10.1016/j.knosys.2025.114724","DOIUrl":"10.1016/j.knosys.2025.114724","url":null,"abstract":"<div><div>High-dimensional psychological data poses challenges due to noise, overlap, and projection distortion. This study presents Bootstrap-Guided Iterative Clustering Refinement (BGICR), a new framework developed to improve clustering in reduced-dimensional spaces. The proposed method uses silhouette-guided filtering and bootstrap sampling to iteratively remove ambiguous points through structural denoising, and it monitors validation scores until convergence. We used real-world psychological assessment data and applied four dimensionality reduction techniques: t-distributed stochastic neighbour embedding, uniform manifold approximation and projection, isometric mapping, and kernel principal component analysis. Results showed that BGICR consistently outperformed conventional clustering pipelines, with uniform manifold approximation and projection yielding the most distinct and well-separated clusters. Through adaptive iterations, the refinement improved the silhouette score to 0.7405, reduced the Davies–Bouldin index to 0.3914, increased the Calinski–Harabasz score to 3755.08, and achieved a Dunn index of 0.7689. Additional validation on synthetic data (Two-Moons) and biomedical datasets (LC25000 histopathological images) confirmed improved clustering quality with stable convergence and efficient runtime. Taken together, these results establish BGICR as a statistically grounded, noise-sensitive, and generalizable method for high-dimensional data analysis across psychological, synthetic, and biomedical domains.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114724"},"PeriodicalIF":7.6,"publicationDate":"2025-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145339765","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":"MLLMs-MR: Multi-modal recognition based on multi-modal large language models","authors":"Shengwei Fu ,&nbsp;Mingyang Yu ,&nbsp;Kaichen OuYang ,&nbsp;Qingsong Fan ,&nbsp;Haisong Huang","doi":"10.1016/j.knosys.2025.114717","DOIUrl":"10.1016/j.knosys.2025.114717","url":null,"abstract":"<div><div>To address the challenges of recognizing data from different modalities, this study proposes a multi-modal recognition method based on Multi-modal Large Language Models (MLLMs-MR), which can process six distinct data types: images, videos, audio, thermal, point cloud, and event data. Existing methods, such as UniBind, treat language as the central modality and construct a text-centric representation space, effectively reducing the representation imbalance among different modalities and improving recognition accuracy. However, descriptions generated by Multi-modal Large Language Models (MLLMs) are directly used for contrastive learning of text embeddings, which would result in a loss of semantic information from the original input. Furthermore, sole reliance on Large Language Models (LLMs) to be used as embedding centers can lead to category misclassification. To address these issues, we have proposed three key improvements based on UniBind: (1) constructing a category-based knowledge base using MLLMs, effectively reducing irrelevant descriptions; (2) designing fusion embedding center localization, which utilizes LLMs, MLLMs, and basic prompts to enhance the robustness of embedding centers; and (3) proposing a cross-modal attention mechanism that incorporates MLLMs-generated descriptions during training, enabling the model to better learn semantic information from multi-modal data and enhance feature representation. Subsequently, MLLMs-enhanced embeddings are aligned with class labels by contrastive learning to enable the recognition of multi-modal data. Experimental results demonstrate that MLLMs-MR outperforms existing models in multi-modal zero-shot recognition, with a 6.42 % accuracy gain on MSR-VTT. It shows an improvement of 8.19 % during fine-tuning on the ESC 5-fold audio dataset.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114717"},"PeriodicalIF":7.6,"publicationDate":"2025-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145363093","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 simple yet effective difficulty-aware bucketed fine-tuning strategy for LLM-based recommendation","authors":"Qianyang Zhu,&nbsp;Bo Yang,&nbsp;Wei Liu,&nbsp;Jiajin Wu","doi":"10.1016/j.knosys.2025.114655","DOIUrl":"10.1016/j.knosys.2025.114655","url":null,"abstract":"<div><div>Recommendation systems have been widely researched and applied in the industry. Recently, with the advancement of large language models (LLMs), there has been much research on LLM-based recommendation models (LLM-RMs), yielding outstanding performance compared with traditional recommendation models. In many LLM-RMs, supervised fine-tuning plays a pivotal role in enhancing model performance. However, existing research primarily focuses on enriching the information of the fine-tuning data, to enable LLMs to learn specific capabilities effectively. In this paper, we study the fine-tuning for LLM-RMs from a different perspective. Specifically, we propose a Difficulty-aware Bucketed Fine-tuning (DBF) strategy to replace the random sampling approach commonly used in existing research. The main philosophy of the proposed DBF strategy is to fine-tune an LLM from easy samples to difficult samples, which simulates the progressive learning process of humans. First, we propose a metric to measure the difficulty of fine-tuning samples based on three aspects: category entropy, category consistency, and category similarity. Based on the proposed metric, we design the bucketed tuning strategy that considers both intra-bucket and inter-bucket difficulty. In addition, we propose a Coarse-to-fine-grained Prompting LLM-based Recommendation Model, CP4Rec, which adopts a two-step reasoning process for making recommendations. We conduct extensive experiments on four real-world benchmark datasets, and results demonstrate that CP4Rec, fine-tuned with the proposed DBF strategy, outperforms the state-of-the-art LLM-RMs across these datasets. Experimental results also highlight the importance of using the buckets in the fine-tuning process to prevent performance degradation. The implementation code is available at <span><span>https://anonymous.4open.science/r/CP4Rec1</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114655"},"PeriodicalIF":7.6,"publicationDate":"2025-10-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145339511","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":"Towards intelligent landslide susceptibility evaluation: Knowledge extraction and rule mining","authors":"Xuexi Yang ,&nbsp;Qian Xu ,&nbsp;Qinghao Liu ,&nbsp;Xin Hu ,&nbsp;Guran Xie ,&nbsp;Yifan Jiang ,&nbsp;Dejin Zhang ,&nbsp;Min Deng","doi":"10.1016/j.knosys.2025.114656","DOIUrl":"10.1016/j.knosys.2025.114656","url":null,"abstract":"<div><div>Landslide susceptibility evaluation (LSE) plays a crucial role in disaster prevention and mitigation. However, current models struggle to achieve an optimal balance among accuracy, interpretability, and scalability. This study proposes a knowledge extraction framework that integrates scientific literature and multi-source spatiotemporal data, aiming to address these limitations by acquiring robust and reliable susceptibility knowledge. First, a landslide susceptibility ontology is designed to systematically organize domain knowledge, encompassing disaster-causing factors, disaster-prone environments, and bearing body attributes. Knowledge extraction employs the improved CasRel model (ERNIE-CasRel) model to derive entity-relationship triples from unstructured literature and then mine knowledge representing expert experience from the triples. Simultaneously, this study integrates self-organizing maps (SOM) and Apriori algorithms to mine spatial aggregation patterns and association rules from structured datasets. The extracted knowledge is then semantically aligned and conflicts are resolved before being integrated into a queryable knowledge graph, which is subsequently stored in Neo4j. Experiments conducted in Yunnan Province, China, validate the efficacy of the proposed framework. Specifically, the ERNIE-CasRel model achieves an F1-score of 0.752 for triple extraction, while the integration of self-organizing maps (SOM) and Apriori algorithms identifies high-confidence association rules. Furthermore, cross-validation leveraging historical landslide data confirms the reliability of these extracted rules. This study advances intelligent landslide susceptibility evaluation (LSE) by synergizing domain knowledge with data-driven techniques, thereby providing a scalable and adaptable solution for geological hazard management. The potential applicability of the proposed methodology to other regions and types of hazards underscores its significant potential for integration into knowledge-based disaster mitigation systems.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114656"},"PeriodicalIF":7.6,"publicationDate":"2025-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145363167","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":"Knowledge-driven deep learning approaches for computer vision tasks: A survey","authors":"Fatima Ezzahra Benkirane,&nbsp;Nathan Crombez,&nbsp;Vincent Hilaire,&nbsp;Yassine Ruichek","doi":"10.1016/j.knosys.2025.114645","DOIUrl":"10.1016/j.knosys.2025.114645","url":null,"abstract":"<div><div>Hybrid artificial intelligence aims to integrate data-driven techniques with knowledge-based systems, offering a promising avenue to enhance artificial intelligence systems accuracy, interoperability, and explainability. Within this domain, neuro-symbolic artificial intelligence represents a sub-field focusing on merging specifically deep neural networks with knowledge-based systems for improved effectiveness. This paper provides a comprehensive overview of recent advancements in the field, specifically focusing on knowledge-driven training approaches for computer vision tasks where knowledge-based systems are deeply integrated into the deep neural networks training process. This integration takes advantage of structured domain knowledge to guide feature extraction. It improves robustness against noisy and incomplete data, allows more reliable and interpretable decision-making mechanisms, and facilitates better generalization in diverse and complex scenarios. These enhancements ultimately improve the overall performance of the neural networks. The presented approaches in this survey are categorized based on the integration level of knowledge within deep neural networks, including input integration, intermediate-level integration, and integration into the loss function. Additionally, the methodologies are sub-categorized based on the knowledge representation extracted from the knowledge-based systems before integration into the deep learning model. The integration methodology for each approach is highlighted to provide a comprehensive comparison between the different contributions. Through a survey of the literature, this paper identifies gaps in understanding the collaboration of knowledge-based systems and deep neural networks in the computer vision field. State-of-the-art approaches are analyzed and compared, evaluating their methodologies, integration knowledge strategy, and application domain. Our work also highlights the strengths and weaknesses of the approaches, discusses the challenges, and provides a critical review of their effectiveness. The paper concludes by exploring potential improvements and outlines future research directions to advance the integration of knowledge-based systems and deep neural networks.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114645"},"PeriodicalIF":7.6,"publicationDate":"2025-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145339505","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":"Enhancing security in IoT-based photovoltaic monitoring with hybrid key management and deep learning optimized routing","authors":"P. Saranya ,&nbsp;R. Rajesh","doi":"10.1016/j.knosys.2025.114652","DOIUrl":"10.1016/j.knosys.2025.114652","url":null,"abstract":"<div><div>The modern power infrastructure faces significant challenges in ensuring reliable and efficient electricity delivery amid rapidly increasing demand across various sectors. This research proposes an Internet of Things (IoT)-based smart grid system integrating Shuffled Frog Leaping Algorithm Optimized Recurrent Neural Network (SFLA-RNN) routing protocol to find the shortest route to reach end user, with Hybrid Paillier Improved Blow Fish (HPIBF) algorithm for key management, adding an extra degree of data protection. The system’s operational status is visualized using the Adafruit IoT dashboard. The validation of developed system is examined using NS2 software and the outcomes reveals superior results with improved Packet Delivery Ratio (PDR) of 98.95%, reduced consumption of energy to 0.024 mJ (100 nodes), longer network lifetime up to 3881 rounds (500 nodes) and minimized latency to 1.6–4.1 s compared to state of art topologies. Moreover, the proposed HPIBF approach achieves encryption and decryption times of 15 ms and 0.35 ms, respectively, outperforming existing algorithms. This confirms that the proposed research on IoT-based monitoring systems lowers operating expenses by improving energy efficiency through the reduction of power loss during transmission and distribution.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114652"},"PeriodicalIF":7.6,"publicationDate":"2025-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145363169","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":"Cross-frequency aware network for camouflaged object detection with octave-transformer","authors":"Feng Dong ,&nbsp;Jinchao Zhu ,&nbsp;Hongpeng Wang","doi":"10.1016/j.knosys.2025.114638","DOIUrl":"10.1016/j.knosys.2025.114638","url":null,"abstract":"<div><div>Camouflaged object detection (COD) aims to identify objects that are fully blended into their surrounding environments. Current mainstream COD methods primarily focus on pixel-level optimization using convolutional neural networks (CNNs), without sufficiently addressing the significance of frequency interactions between candidate targets and noisy backgrounds, which are crucial for obtaining accurate edge and localization information. This paper explores the integration of multi-frequency features, and constructs a cross-frequency aware network (CFANet). The proposed network utilizes precisely learned deep-layer low-frequency features to guide other layers, achieving coarse localization. To further refine segmentation, the network employs both Transformer and CNN structures to facilitate the interaction and optimization of high- and low-frequency features at local and global levels. The model adopts a localization-guided decoder structure (LGS) that allows deep-layer low-frequency features to play a key role in guiding localization. The discussion module (DM) comprises three feature extraction experts, who engage in a teacher-student learning framework to derive more accurate deep-layer low-frequency features. In the Octave-Transformer module (OTM), the high- and low-frequency fused features based on octave convolution (OctConv) and Transformer deeply mine semantic features and detailed information. Compared to 33 existing state-of-the-art COD methods, the proposed network achieves overall superior performance across four benchmark datasets. Additionally, the network demonstrates excellent performance in other downstream tasks, such as polyp segmentation, surface defect detection. Our code is available at <span><span>https://github.com/wkkwll-df/CFANet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114638"},"PeriodicalIF":7.6,"publicationDate":"2025-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145363170","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":"Fine-grained label propagation via density-based prototype matching for cross-subject EEG emotion recognition","authors":"Qiang Wang,&nbsp;Liying Yang,&nbsp;Qian Zhang,&nbsp;Jingtao Du,&nbsp;Yumeng Ye","doi":"10.1016/j.knosys.2025.114650","DOIUrl":"10.1016/j.knosys.2025.114650","url":null,"abstract":"<div><div>Emotion recognition based on electroencephalography (EEG) signals has become a prominent research focus in affective computing. However, challenges such as individual differences and label noise have significantly impeded the generalization and accuracy of models. To address these challenges, this study proposes a novel fine-grained label propagation framework based on Density-Based Prototype Matching (DBPM). By leveraging density-based clustering to capture fine-grained subdomain structures, the framework enables robust prototype matching and reliable label propagation across domains. Furthermore, a Sequential Multi-Source Training Strategy is devised to progressively incorporate multiple source domains, thereby ensuring stable one-to-one prototype matching and mitigating inter-source interference. Extensive experiments are conducted on two publicly available EEG emotion datasets (SEED and SEED-IV) under a leave-one-subject-out cross-validation evaluation protocol. The results demonstrate that the proposed DBPM achieves state-of-the-art performance, offering a promising solution for addressing individual differences and label noise in EEG emotion recognition. The source code is publicly available at: <span><span>https://github.com/qwangwl/DBPM</span><svg><path></path></svg></span></div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114650"},"PeriodicalIF":7.6,"publicationDate":"2025-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145363322","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":"Enhancing multi-agent reinforcement learning via world model assisted single-agent population policies in multi-UAV cooperative-competitive scenario","authors":"Jiaming Cheng ,&nbsp;Ni Li ,&nbsp;Changyin Dong ,&nbsp;Chong Tang","doi":"10.1016/j.knosys.2025.114534","DOIUrl":"10.1016/j.knosys.2025.114534","url":null,"abstract":"<div><div>Multi-agent reinforcement learning (MARL) with limited search is prone to becoming trapped in local optima and struggles to adapt to opponents’ changing intentions and strategies in cooperative-competitive environments. Evolutionary strategies (ES) are a promising alternative that have been applied to RL owing to their diverse exploration characteristics to form a hybrid framework. However, most such methods require population policies to interact within the environment for Monte Carlo (MC) evaluation or to filter experience using Q-functions, leading to low sample efficiency. Additionally, existing methods struggle to maintain policy quality and avoid catastrophic forgetting while exploring. Our objective is to enhance MARL in mixed environments through single-agent population policies to improve the sample efficiency and mitigate the aforementioned issues with ES. To achieve this, we propose a world-model-assisted cross entropy method (CEM)-MARL approach. This world model enables inferring opponents’ mental states and predicting and evaluating future trajectories. CEM is used to update population policies and retain the top K/2 old policies. Subsequently, predicted future experiences are reused for updating the MARL policy and evaluated to perform quality-assured mutations on the top K/2 population policies. Furthermore, the elite policy guides the MARL, and the MARL policy guides the population. Experimental results in multiple unmanned aerial vehicle (multi-UAV) game scenarios show that our method accelerates learning by 40 times compared to model-free MARL and nearly 10 times compared to our version without ES. This increases the learning efficiency and enhances the performance of the hybrid framework that incorporates collaborative learning and evolutionary processes.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114534"},"PeriodicalIF":7.6,"publicationDate":"2025-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145363091","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}