Knowledge-Based Systems最新文献

{"title":"Learning hierarchical scene graph and contrastive learning for object goal navigation","authors":"Jian Luo ,&nbsp;Jian Zhang ,&nbsp;Bo Cai ,&nbsp;Yaoxiang Yu ,&nbsp;Aihua Ke","doi":"10.1016/j.knosys.2025.113532","DOIUrl":"10.1016/j.knosys.2025.113532","url":null,"abstract":"<div><div>The task of object goal navigation (ObjNav) requires the agent to locate the given target object within a complex dynamic scene. To successfully accomplish the task, the agent needs to well understand the scenes, make executable decisions with less steps, avoid collisions, and successfully navigate to the target. As a result, efficient environmental perception and scene graph-inspired path planning is important to successfully accomplish the ObjNav task. In this paper, we present a hierarchical scene graph (HSG) contrastive learning, which consists of (1) a multimodal graph mixer that aligns the visual and textual information using open-vocabulary detector with GLIP. It can be regarded as an “eagle eye” to perceive target-related frontiers and suppress irrelevant information, and (2) a graph constructer that takes observed RGBD images to incrementally build a hierarchical scene graph. It acts as the “brain” that memorizes the common scene layout, (3) an action control contrastive learning that takes the graph contextual relationships as input to predict optimal actions to the target. It is treated as the “limbs” of the agent, coordinating and correcting incorrect movements. On the task of ObjNav, experiments on Gibson, HM3D, MP3D, and ProcTHOR demonstrate that navigation plans from the HSG framework achieve significantly higher success rates than existing map-based method, indicating the feasibility of executing navigation utilizing commonsense knowledge from language models leading efficient semantic exploration. <em>Code is available at</em> <span><span>https://github.com/luosword/HSG4VN</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"319 ","pages":"Article 113532"},"PeriodicalIF":7.2,"publicationDate":"2025-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143895417","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":"Adaptive frequency-domain enhanced deep model driven by heterogeneous networks for medical image segmentation","authors":"Dong Liu ,&nbsp;Jin Kuang","doi":"10.1016/j.knosys.2025.113599","DOIUrl":"10.1016/j.knosys.2025.113599","url":null,"abstract":"<div><div>Accurate medical image segmentation necessitates precise localization of global structures and local boundaries due to the high variability in lesion shapes and sizes. However, existing models are limited by conventional spatiotemporal features and single-network architectures, which restrict the simultaneous captures of semantic information and boundary details, thereby challenging generalizable medical image segmentation. To overcome these limitations, we propose a heterogeneous network-driven adaptive frequency-domain enhanced deep model(AFDSeg). First, we introduce the Frequency Domain Adaptive High-Frequency Feature Selection(FAHS) module, which adaptively extracts high-frequency features to enhance contour and detail representation while integrating spatiotemporal and frequency-domain features for improved consistency. Additionally, Prototype-Guided Low-Frequency Feature Aware(PFLA) and Local High-Frequency Salient-Feature Denoising (LHSD) modules are developed, which extract discriminative low-frequency features while suppressing local noise in high-frequency components, thereby facilitating efficient multi-scale feature fusion. Furthermore, the Multi-Level Prototype Feature Refinement(MPFR) Module is introduced to align low- and high-dimensional features during decoding and enhance semantic consistency. Finally, a heterogeneous network framework capable of accommodating multiple network architecture for medical image segmentation is proposed. Our method achieves mDice scores of 93.91%, 88.64%, 90.70%, 91.27%, and 81.38% on the Kvasir-SEG, BUSI, ISIC-2017, ACDC, and Synapse datasets, respectively, and attains 92.09%, 93.50%, and 83.92% in cross-domain experiments on three unseen datasets (Kvasir Capsule-SEG, BUS42, and M&amp;Ms). Our approach consistently outperforms state-of-the-art methods on both benchmark and cross-domain datasets. Extensive quantitative and qualitative experiments demonstrated that AFDSeg accurately segments global structures and local details while maintaining superior generalization, underscoring its clinical significance. The Code is available at <span><span>https://github.com/promisedong/AFDSeg</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"319 ","pages":"Article 113599"},"PeriodicalIF":7.2,"publicationDate":"2025-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143899626","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":"Improved proximal policy optimization for UAV tracking in complex environments","authors":"Tao Zhang ,&nbsp;Qingyan Zhou ,&nbsp;Yue Zheng ,&nbsp;Huiwen Yu","doi":"10.1016/j.knosys.2025.113627","DOIUrl":"10.1016/j.knosys.2025.113627","url":null,"abstract":"<div><div>Unmanned Aerial Vehicles (UAVs) operating in urban environments face critical challenges in dynamic field of view (FOV) management and obstacle avoidance. To address these issues, this paper proposes an improved Proximal Policy Optimization algorithm (I-PPO) that integrates seven key enhancements, including reward scaling, gradient clip, and others. This algorithm improves sample efficiency and reduces policy oscillation in complex environments, in which we have developed a three-dimensional simulation environment capable of multi-terrain parametric modeling that integrates weather-related FOV attenuation models and intelligent dynamic obstacle modules. Focusing on the tracking task, the study designs a reward function based on a hierarchical penalty system and priority rules. This approach ensures operational safety while maximizing target vehicle visibility, thereby optimizing agent performance under environmental uncertainties. Experimental results demonstrate that in plain environments, I-PPO yields a 2.9-fold increase in mean cumulative reward and extends target tracking duration by a factor of 2.7 compared to the standard PPO. In hilly terrain, I-PPO maintains reward performance comparable to its plain environment baseline, exhibiting merely a 2% performance degradation, confirming terrain adaptability. In mountainous terrain, while it shows a 12% reward reduction versus hilly terrain, it exhibits a 38.9% reduction in reward variance (measured by IQR) compared to Discrete Soft Actor–Critic (DSAC), this demonstrates significant robustness enhancement. In scenarios with 10 intelligent dynamic obstacles, the algorithm achieves stable convergence within 984 time units and demonstrates equivalent robustness under weather-induced FOV attenuation across multi-terrain environments. Furthermore, Theoretical analysis confirms the method’s compliance with policy gradient convergence requirements.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"319 ","pages":"Article 113627"},"PeriodicalIF":7.2,"publicationDate":"2025-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143899627","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":"Test-driving information theory-based compositional distributional semantics: A case study on Spanish song lyrics","authors":"Adrián Ghajari ,&nbsp;Alejandro Benito-Santos ,&nbsp;Salvador Ros ,&nbsp;Víctor Fresno ,&nbsp;Elena González-Blanco","doi":"10.1016/j.knosys.2025.113549","DOIUrl":"10.1016/j.knosys.2025.113549","url":null,"abstract":"<div><div>Song lyrics pose unique challenges for semantic similarity assessment due to their metaphorical language, structural patterns, and cultural nuances - characteristics that often challenge standard natural language processing (NLP) approaches. These challenges stem from a tension between compositional and distributional semantics: while lyrics follow compositional structures, their meaning depends heavily on context and interpretation. The Information Theory-based Compositional Distributional Semantics framework offers a principled approach by integrating information theory with compositional rules and distributional representations. We evaluate eight embedding models on Spanish song lyrics, including multilingual, monolingual contextual, and static embeddings. Results show that multilingual models consistently outperform monolingual alternatives, with the domain-adapted ALBERTI achieving the highest F1 macro scores (78.92 ± 10.86). Our analysis reveals that monolingual models generate highly anisotropic embedding spaces, significantly impacting performance with traditional metrics. The Information Contrast Model metric proves particularly effective, providing improvements up to 18.04 percentage points over cosine similarity. Additionally, composition functions maintaining longer accumulated vector norms consistently outperform standard averaging approaches. Our findings have important implications for NLP applications and challenge standard practices in similarity calculation, showing that effectiveness varies with both task nature and model characteristics.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"319 ","pages":"Article 113549"},"PeriodicalIF":7.2,"publicationDate":"2025-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143913261","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":"Personalized e-learning resource recommendation using multimodal-enhanced collaborative filtering","authors":"Xinwei Zhai ,&nbsp;Yuanyuan Wang ,&nbsp;Luwen Liang ,&nbsp;Kangzhong Wang ,&nbsp;Fengchun Pei ,&nbsp;Eugene Yujun Fu","doi":"10.1016/j.knosys.2025.113605","DOIUrl":"10.1016/j.knosys.2025.113605","url":null,"abstract":"<div><div>Personalized learning resource recommendation is a prominent research area in the field of e-learning, allowing learners to find appropriate resources that align with their specific learning needs. The continuous development and optimization of online learning platforms have resulted in an increasing amount of e-learning resources and learner data. This poses challenges to the existing e-learning resource recommendation approaches, most of which rely on conventional collaborative filtering (CF) exclusively. Their efficiency is constrained owing to the utilization of a sole modality or a limited subset of modalities for the recommendation. To address these challenges, this study proposes a multimodal-enhanced CF approach in e-learning. Our approach uses various modalities for modeling, including learners’ learning records, human–computer interaction patterns, and information related to the resources. It integrates techniques such as matrix factorization for the joint learner–resource pattern modeling, clustering for grouping similar learners, and the long short-term memory network for capturing the temporal dynamics of learning activities. Comprehensive experiments are conducted to evaluate the efficiency of the proposed approach, and to determine its optimal setup for a deep understanding of the contributions of each component.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"319 ","pages":"Article 113605"},"PeriodicalIF":7.2,"publicationDate":"2025-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143895418","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 gate-enhanced neuro additive graph neural network via knowledge distillation for CTR prediction","authors":"Fei Guan ,&nbsp;Jing Yang ,&nbsp;Chenxia Jin","doi":"10.1016/j.knosys.2025.113668","DOIUrl":"10.1016/j.knosys.2025.113668","url":null,"abstract":"<div><div>The click-through rate (CTR) prediction is a crucial task in commercial recommender systems and online advertising platforms. Recent studies have revealed shortcomings in CTR optimization, particularly in their ability to effectively identify and interpret the abnormal or latent feature interactions obscured by user behaviors. In this paper, a novel CTR prediction model is developed in two stages. The first stage formulates a gate-enhanced neuro additive graph neural network (GNAGNN), by dynamically capturing the complex interactions between features in different input environments, the adaptability of the model to the importance of features is significantly improved. While the second stage utilizes the knowledge distillation framework, enabling GNAGNN to effectively learn from a gated ensemble of existing CTR models. Unlike most higher-order feature interaction models that rely on deep neural networks, our method avoids high-complexity matrix computation and significantly reduces the computational overhead. Specifically, the framework adopts a dynamic parametric mechanism to determine the weight of the model involved in the prediction through the continuous action vector, and then achieve the accurate prediction of each advertisement impression. Eventually, comprehensive experiments carried out on two public datasets convincingly demonstrate that GNAGNN significantly outperforms the state-of-the-art baselines, and it can offer precise and interpretable understandings into features and their interactions while reducing computational costs.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"319 ","pages":"Article 113668"},"PeriodicalIF":7.2,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143906447","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":"Research on knowledge drift based on interaction matching","authors":"Yuanjian Lin ,&nbsp;Yongwang Duan ,&nbsp;Chenxia Jin ,&nbsp;Fachao Li","doi":"10.1016/j.knosys.2025.113664","DOIUrl":"10.1016/j.knosys.2025.113664","url":null,"abstract":"<div><div>Drift detection is a fundamentally challenging task in data-driven decision-making; however, traditional distribution-based methods for data drift detection fail to adequately capture the characteristics of knowledge-centric data-driven decision-making. Thus, exploring knowledge-based data drift detection methods has extensive practical value. Herein, we consider If-Then rules as the knowledge representation framework, focusing on changes in the quantity and quality of knowledge and utilizing the interaction matching of rules as a measure of knowledge differentiation. First, we propose the concept of forward and reverse rule (trust) drift and introduce a knowledge drift detection model based on interaction matching (named rule knowledge drift detection [RKDD]). The basic features of RKDD are analyzed and discussed. Second, using statistical theory, we discuss the convergence characteristics of rule knowledge and propose a knowledge drift detection model that utilizes interaction matching within the framework of sampling (named sample-rule knowledge drift detection [S-RKDD]). Finally, we compare and analyze the effectiveness of RKDD using three University of California Irvine (UCI) datasets. Theoretical analysis and experimental results demonstrate that RKDD possesses good structural characteristics and interpretability, enabling the integration of decision awareness into the decision-making process through simple parameter adjustments, thereby enriching the existing data drift detection theory to a certain extent.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"319 ","pages":"Article 113664"},"PeriodicalIF":7.2,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143899573","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":"Multi-domain fake news detection method based on generative adversarial network and graph network","authors":"Xuefeng Li ,&nbsp;Jian Wei ,&nbsp;Chensu Zhao ,&nbsp;Xiaqiong Fan ,&nbsp;Yuhang Wang","doi":"10.1016/j.knosys.2025.113665","DOIUrl":"10.1016/j.knosys.2025.113665","url":null,"abstract":"<div><div>The proliferation of misinformation in today's digital era poses significant challenges, with fake news detection becoming critical to mitigate economic losses and social instability. Despite extensive research efforts, most existing approaches are tailored for single-domain fake news detection, struggling with data distribution discrepancies and domain shifts when applied to multi-domain scenarios. This limitation underscores the urgent need for solutions that address the complexities of cross-domain detection. Here, we propose a novel framework MFGAG that synergistically integrates adversarial networks and graph neural networks with emotional, stylistic, and semantic features to enable precise domain localization. By leveraging these features, the framework effectively models intricate relationships among news articles within the same temporal context, addressing the challenges posed by multi-domain datasets. Experimental evaluations demonstrate that our approach outperforms state-of-the-art methods, achieving an average accuracy improvement of 3.3 percentage points for single-domain news and nearly 1 percentage point for mixed-domain data, culminating in an overall accuracy of 93.1 %. The code involved in this study is publicly available on website <span><span>https://github.com/SWLee777/MFGAG</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"319 ","pages":"Article 113665"},"PeriodicalIF":7.2,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143895419","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":"Reinforcement and opposition-based learning enhanced weighted mean of vectors algorithm for global optimization and feature selection","authors":"İlker Gölcük ,&nbsp;Fehmi Burcin Ozsoydan ,&nbsp;Esra Duygu Durmaz","doi":"10.1016/j.knosys.2025.113626","DOIUrl":"10.1016/j.knosys.2025.113626","url":null,"abstract":"<div><div>This paper presents a novel optimization algorithm that integrates reinforcement learning (RL) and opposition-based learning (OBL) mechanisms with the weighted mean of vectors algorithm (INFO). The OBL has proven effective in enhancing optimization algorithms, the lack of adaptive selection mechanisms often leads to suboptimal performance. The proliferation of OBL variants poses significant challenges in selecting appropriate mechanisms for specific optimization problems, as each variant exhibits distinct characteristics and performance patterns across different problem landscapes. This research addresses this limitation by introducing a novel RL framework for OBL selection. The proposed QLOBL_INFO algorithm employs Q-learning to adaptively select among five OBL variants, enabling dynamic strategy adaptation during the optimization process. The algorithm's performance has been extensively evaluated using the CEC2022 benchmark suite, real-world feature selection problems, and constrained optimization problems. These results demonstrate that RL-based adaptive OBL selection represents an effective approach for enhancing optimization performance, particularly in complex optimization landscapes and real-world applications.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"319 ","pages":"Article 113626"},"PeriodicalIF":7.2,"publicationDate":"2025-04-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143903247","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":"From local verification to global reasoning: Exploiting slot-accompanying update for improved slot selection","authors":"Bing Qian ,&nbsp;Jinyu Guo ,&nbsp;Qiwei Wang ,&nbsp;Kai Shuang","doi":"10.1016/j.knosys.2025.113521","DOIUrl":"10.1016/j.knosys.2025.113521","url":null,"abstract":"<div><div>The goal of dialogue-state tracking (DST) is to determine the current state of a dialogue by analysing the entire preceding dialogue context. Nonetheless, current approaches frequently fail to account for the significance of concurrent updates, where related slots must be updated simultaneously based on their historical relationships, even in the absence of explicit signals in the current dialogue turn. To address this limitation, we introduce From Local Verification to Global Reasoning (FLV2GR), an innovative method that improves slot-update selection by combining local verification of present dialogue details with global reasoning over historical dialogue data. Our approach utilizes a graph neural network (GNN) to model and infer interdependencies between slots, enabling the identification of accompanying update relationships that are frequently overlooked by other approaches. This comprehensive selection mechanism improves the precision of slot updates, thereby enhancing overall DST performance. The FLV2GR model establishes a new performance benchmark on the MultiWOZ 2.1, 2.2, and 2.4 datasets, showcasing its effectiveness in capturing both local and global dialogue dynamics for more precise and reliable DST.<span><span>¹</span></span></div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"319 ","pages":"Article 113521"},"PeriodicalIF":7.2,"publicationDate":"2025-04-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143886860","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}