Knowledge-Based Systems最新文献

英文中文

Personalized recommendation by integrating a neural topic model and Bayesian personalized ranking

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-06 DOI: 10.1016/j.knosys.2025.113110

Yixin Zhang , Sichen Lin , Zhili Zhao , Xuran Zhu , Chenbo He

Traditional recommendation systems typically sort items and provide users with top items by analyzing user–item interactions. Interactions vary from person to person because they are determined by personal intents and other environmental factors. However, these intents are implicit and difficult to capture because experimental data often contain plain user–item interactions. In this study, we proposed an attentive neural topic model (ANTM) to determine user latent intents and distinguish individual preferences. We first used the neural topic model in the natural language processing domain to discover user latent intents by encoding user–item interactions and jointly learned the model and variational parameters during inference. In addition, because of differences in user latent intents, we applied an attention mechanism to intents to obtain individual preferences. The representation of user features enriched by individual latent intents was then used to replace plain user profiles to provide personalized recommendations. Experimental results demonstrated that the proposed ANTM outperformed the best baseline algorithm by 1.09%–17.25% and 0.66%–10.38% in terms of the hit rate for recommending the top-5 and top-10 items, respectively. Moreover, its improvements over the best baseline algorithm were 0.69%–35.48% and 0.54%–15.48% in terms of normalized discounted cumulative gain in recommending the top-5 and top-10 items, respectively.

{"title":"Personalized recommendation by integrating a neural topic model and Bayesian personalized ranking","authors":"Yixin Zhang , Sichen Lin , Zhili Zhao , Xuran Zhu , Chenbo He","doi":"10.1016/j.knosys.2025.113110","DOIUrl":"10.1016/j.knosys.2025.113110","url":null,"abstract":"<div><div>Traditional recommendation systems typically sort items and provide users with top items by analyzing user–item interactions. Interactions vary from person to person because they are determined by personal intents and other environmental factors. However, these intents are implicit and difficult to capture because experimental data often contain plain user–item interactions. In this study, we proposed an attentive neural topic model (<em>ANTM</em>) to determine user latent intents and distinguish individual preferences. We first used the neural topic model in the natural language processing domain to discover user latent intents by encoding user–item interactions and jointly learned the model and variational parameters during inference. In addition, because of differences in user latent intents, we applied an attention mechanism to intents to obtain individual preferences. The representation of user features enriched by individual latent intents was then used to replace plain user profiles to provide personalized recommendations. Experimental results demonstrated that the proposed <em>ANTM</em> outperformed the best baseline algorithm by 1.09%–17.25% and 0.66%–10.38% in terms of the hit rate for recommending the top-5 and top-10 items, respectively. Moreover, its improvements over the best baseline algorithm were 0.69%–35.48% and 0.54%–15.48% in terms of normalized discounted cumulative gain in recommending the top-5 and top-10 items, respectively.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"311 ","pages":"Article 113110"},"PeriodicalIF":7.2,"publicationDate":"2025-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143377619","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}

引用次数: 0

Structured reasoning and answer verification: Enhancing question answering system accuracy and explainability

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-05 DOI: 10.1016/j.knosys.2025.113091

Jihyung Lee , Gary Geunbae Lee

The performance of question-answering (QA) models has significantly advanced, yet challenges remain in verifying the accuracy of generated answers and providing clear explanations of the reasoning behind them. In response, this study introduces a novel answer verification model that detects inaccuracies in QA system outputs and offers structured, multi-step explanations to enhance both understanding and reliability. We built an answer verification system consisting of a stepwise prover and two types of verifiers and tested the proposed system on the EntailmentBank dataset as well as the ARC, AQUA-RAT, and AR-LSAT datasets from the STREET benchmark. By correcting the answers generated by the T5-large and GPT-3.5 QA models and comparing the results before and after correction, we observed notable improvements in answer accuracy and explanation clarity. Specifically, the proposed model increased the exact match score of the T5-large model by 1.76% and that of GPT-3.5 by 3.53% on the EntailmentBank dataset. Additionally, to address potential data scarcity, the study proposes a data augmentation technique that employs large language models and multi-hop datasets to generate reasoning chains, thereby enriching the training data. Although the augmented data did not match the quality of the gold data, which is manually curated and verified by humans, our experiments demonstrated that combining gold data with augmented data resulted in better performance than using only a subset of the gold data.

{"title":"Structured reasoning and answer verification: Enhancing question answering system accuracy and explainability","authors":"Jihyung Lee , Gary Geunbae Lee","doi":"10.1016/j.knosys.2025.113091","DOIUrl":"10.1016/j.knosys.2025.113091","url":null,"abstract":"<div><div>The performance of question-answering (QA) models has significantly advanced, yet challenges remain in verifying the accuracy of generated answers and providing clear explanations of the reasoning behind them. In response, this study introduces a novel answer verification model that detects inaccuracies in QA system outputs and offers structured, multi-step explanations to enhance both understanding and reliability. We built an answer verification system consisting of a stepwise prover and two types of verifiers and tested the proposed system on the EntailmentBank dataset as well as the ARC, AQUA-RAT, and AR-LSAT datasets from the STREET benchmark. By correcting the answers generated by the T5-large and GPT-3.5 QA models and comparing the results before and after correction, we observed notable improvements in answer accuracy and explanation clarity. Specifically, the proposed model increased the exact match score of the T5-large model by 1.76% and that of GPT-3.5 by 3.53% on the EntailmentBank dataset. Additionally, to address potential data scarcity, the study proposes a data augmentation technique that employs large language models and multi-hop datasets to generate reasoning chains, thereby enriching the training data. Although the augmented data did not match the quality of the gold data, which is manually curated and verified by humans, our experiments demonstrated that combining gold data with augmented data resulted in better performance than using only a subset of the gold data.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"311 ","pages":"Article 113091"},"PeriodicalIF":7.2,"publicationDate":"2025-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143372720","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}

引用次数: 0

IReGNN: Implicit review-enhanced graph neural network for explainable recommendation

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-05 DOI: 10.1016/j.knosys.2025.113113

Qingbo Hao , Chundong Wang , Yingyuan Xiao , Wenguang Zheng

Explainable recommendations can not only recommend items to users but also provide corresponding explanations, which is crucial for enhancing the transparency, credibility, and security of the system. Reviews, as an important information source for explainable recommendations, have received considerable attention. However, existing review-based explainable recommendations focus primarily on exploring user preferences and item features, as well as generating explanations from reviews, overlooking the limitations imposed by review sparsity on model performance. To address this issue, we propose an Implicit Review-enhanced Graph Neural Network (IReGNN) for explainable recommendations. Specifically, we construct a review network and a rating network, respectively. For the review network, we adopt an unsupervised approach to mine different topics of users and items, thereby enhancing node attribute representations. On the other hand, for the rating network, we extract implicit relationships between individuals and generate virtual reviews under the constraint of topics, which can effectively alleviate the data sparsity issue. Finally, we leverage a spatial graph neural network to learn node representations, generating accurate recommendations and high-quality explanations. Through a series of experiments on three publicly available datasets, results demonstrate that IReGNN outperforms eight baseline models in terms of rating prediction and explanation quality. Moreover, our model also has certain advantages in sparse data scenarios. The model and datasets are released at: https://github.com/SamuelZack/IReGNN.git.

{"title":"IReGNN: Implicit review-enhanced graph neural network for explainable recommendation","authors":"Qingbo Hao , Chundong Wang , Yingyuan Xiao , Wenguang Zheng","doi":"10.1016/j.knosys.2025.113113","DOIUrl":"10.1016/j.knosys.2025.113113","url":null,"abstract":"<div><div>Explainable recommendations can not only recommend items to users but also provide corresponding explanations, which is crucial for enhancing the transparency, credibility, and security of the system. Reviews, as an important information source for explainable recommendations, have received considerable attention. However, existing review-based explainable recommendations focus primarily on exploring user preferences and item features, as well as generating explanations from reviews, overlooking the limitations imposed by review sparsity on model performance. To address this issue, we propose an Implicit Review-enhanced Graph Neural Network (IReGNN) for explainable recommendations. Specifically, we construct a review network and a rating network, respectively. For the review network, we adopt an unsupervised approach to mine different topics of users and items, thereby enhancing node attribute representations. On the other hand, for the rating network, we extract implicit relationships between individuals and generate virtual reviews under the constraint of topics, which can effectively alleviate the data sparsity issue. Finally, we leverage a spatial graph neural network to learn node representations, generating accurate recommendations and high-quality explanations. Through a series of experiments on three publicly available datasets, results demonstrate that IReGNN outperforms eight baseline models in terms of rating prediction and explanation quality. Moreover, our model also has certain advantages in sparse data scenarios. The model and datasets are released at: <span><span>https://github.com/SamuelZack/IReGNN.git</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"311 ","pages":"Article 113113"},"PeriodicalIF":7.2,"publicationDate":"2025-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143372986","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}

引用次数: 0

Spatiotemporal interactive learning dynamic adaptive graph convolutional network for traffic forecasting

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-05 DOI: 10.1016/j.knosys.2025.113115

Feng Jiang , Xingyu Han , Shiping Wen , Tianhai Tian

Traffic forecasting plays a critical role in tasks such as route planning and traffic management. Recent advancements in graph neural networks have enabled the effective modeling of spatiotemporal correlations, significantly enhancing traffic prediction accuracy. However, most existing research primarily focuses on general spatiotemporal characteristics shared across all nodes, often neglecting the unique attributes of individual nodes. Additionally, these studies tend to overlook the diverse temporal features inherent in the data, limiting their ability to fully capture complex spatiotemporal dependencies. To tackle these challenges, this study introduces the Spatiotemporal Interactive Learning Dynamic Adaptive Graph Convolutional Network (SILDAGCN) for traffic forecasting. Specifically, SILDAGCN incorporates a data embedding module to integrate temporal features into the raw data and extract critical information effectively. Moreover, it employs a dynamic adaptive graph convolutional network designed to capture real-time spatiotemporal dynamics and uncover both shared and node-specific spatiotemporal correlations. This paper also introduces a spatiotemporal feature interaction learning mechanism designed to capture and learn the diverse, evolving characteristics of spatiotemporal dependencies, enabling mutual enhancement through effective feedback. Finally, the output block leverages convolutional operations to enhance the model’s information extraction capabilities, producing the final traffic network forecasts. Experimental evaluations on four real-world datasets demonstrate that SILDAGCN achieves accurate traffic flow and demand predictions with relatively low computational cost.

{"title":"Spatiotemporal interactive learning dynamic adaptive graph convolutional network for traffic forecasting","authors":"Feng Jiang , Xingyu Han , Shiping Wen , Tianhai Tian","doi":"10.1016/j.knosys.2025.113115","DOIUrl":"10.1016/j.knosys.2025.113115","url":null,"abstract":"<div><div>Traffic forecasting plays a critical role in tasks such as route planning and traffic management. Recent advancements in graph neural networks have enabled the effective modeling of spatiotemporal correlations, significantly enhancing traffic prediction accuracy. However, most existing research primarily focuses on general spatiotemporal characteristics shared across all nodes, often neglecting the unique attributes of individual nodes. Additionally, these studies tend to overlook the diverse temporal features inherent in the data, limiting their ability to fully capture complex spatiotemporal dependencies. To tackle these challenges, this study introduces the Spatiotemporal Interactive Learning Dynamic Adaptive Graph Convolutional Network (SILDAGCN) for traffic forecasting. Specifically, SILDAGCN incorporates a data embedding module to integrate temporal features into the raw data and extract critical information effectively. Moreover, it employs a dynamic adaptive graph convolutional network designed to capture real-time spatiotemporal dynamics and uncover both shared and node-specific spatiotemporal correlations. This paper also introduces a spatiotemporal feature interaction learning mechanism designed to capture and learn the diverse, evolving characteristics of spatiotemporal dependencies, enabling mutual enhancement through effective feedback. Finally, the output block leverages convolutional operations to enhance the model’s information extraction capabilities, producing the final traffic network forecasts. Experimental evaluations on four real-world datasets demonstrate that SILDAGCN achieves accurate traffic flow and demand predictions with relatively low computational cost.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"311 ","pages":"Article 113115"},"PeriodicalIF":7.2,"publicationDate":"2025-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143340038","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}

引用次数: 0

Multi-dimension rotations based on quaternion system for modeling various patterns in temporal knowledge graphs

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-05 DOI: 10.1016/j.knosys.2025.113114

Jun Zhu , Jiahui Hu , Di Bai , Yan Fu , Junlin Zhou , Duanbing Chen

Missing information is a prevalent occurrence in temporal knowledge graphs (TKGs), and thus, TKG completion holds considerable importance. Modeling the diverse relational patterns inherent in TKGs is crucial for this process. However, existing methods mainly focus on pre-existing patterns within knowledge graphs while neglecting the influence of temporal information. It is common for multiple relationships to exist between two entities at the same moment, as well as for the same event to transpire at different timestamps. Existing models primarily rely on single or sequential transformations, rendering them inadequate for modeling these intricate patterns. To tackle these challenges, we propose a novel model, multi-dimension rotations based on quaternion system (MDRQS), that integrates the attention mechanism to fuse rotations of different dimensions for modeling interactions between entities. This complex combination of transformations, utilizing parallelization, enables the modeling of the aforementioned patterns through multi-dimensional rotations. The attention mechanism determines the most appropriate dimensional rotation for different facts at various timestamps. We demonstrate that MDRQS effectively models pre-existing and new patterns. Through experiments conducted on four benchmark datasets, the effectiveness of our model is shown in the link prediction task.

{"title":"Multi-dimension rotations based on quaternion system for modeling various patterns in temporal knowledge graphs","authors":"Jun Zhu , Jiahui Hu , Di Bai , Yan Fu , Junlin Zhou , Duanbing Chen","doi":"10.1016/j.knosys.2025.113114","DOIUrl":"10.1016/j.knosys.2025.113114","url":null,"abstract":"<div><div>Missing information is a prevalent occurrence in temporal knowledge graphs (TKGs), and thus, TKG completion holds considerable importance. Modeling the diverse relational patterns inherent in TKGs is crucial for this process. However, existing methods mainly focus on pre-existing patterns within knowledge graphs while neglecting the influence of temporal information. It is common for multiple relationships to exist between two entities at the same moment, as well as for the same event to transpire at different timestamps. Existing models primarily rely on single or sequential transformations, rendering them inadequate for modeling these intricate patterns. To tackle these challenges, we propose a novel model, multi-dimension rotations based on quaternion system (MDRQS), that integrates the attention mechanism to fuse rotations of different dimensions for modeling interactions between entities. This complex combination of transformations, utilizing parallelization, enables the modeling of the aforementioned patterns through multi-dimensional rotations. The attention mechanism determines the most appropriate dimensional rotation for different facts at various timestamps. We demonstrate that MDRQS effectively models pre-existing and new patterns. Through experiments conducted on four benchmark datasets, the effectiveness of our model is shown in the link prediction task.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"311 ","pages":"Article 113114"},"PeriodicalIF":7.2,"publicationDate":"2025-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143349001","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}

引用次数: 0

Interval type-2 fuzzy neural networks for multi-label classification

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-05 DOI: 10.1016/j.knosys.2025.113014

Dayong Tian , Feifei Li , Yiwen Wei

Prediction of multi-dimensional labels plays an important role in machine learning problems. We discovered that traditional binary labels could not capture the contents and relationships in an instance. Hence, we propose a multi-label classification model based on interval type-2 fuzzy logic.

In the proposed model, we use a deep neural network to predict an instance’s type-1 fuzzy membership and another to predict the membership’s fuzzifiers, resulting in interval type-2 fuzzy memberships. We also propose a loss function for determining the similarities between binary labels in datasets and interval type-2 fuzzy memberships generated by our model. The experiments validate that our approach outperforms baselines in multi-label classification benchmarks.

引用次数: 0

Knowledge guided deep deterministic policy gradient

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-04 DOI: 10.1016/j.knosys.2025.113087

Peng Qin, Tao Zhao

Deep deterministic policy gradient (DDPG) exhibits excellent handling capabilities for complex regulation and control problems with continuous state and action spaces. However, its trial-and-error interaction and learning from scratch require extensive exploration by the agent, leading to low learning efficiency and even non-convergence in sparse reward environments. To fully utilize knowledge during the learning process to improve efficiency and performance, this paper draws inspiration from human learning methods and proposes a semantic knowledge-guided DDPG (KGDDPG) approach. In terms of knowledge representation, considering the fuzziness and precision of semantic knowledge, a knowledge system based on a rule framework combining precise propositions and fuzzy propositions is constructed. In terms of knowledge integration, to reduce the randomness of exploration, a knowledge-guided action strategy based on stacked generalization is proposed. Furthermore, a supervised-then-reinforced learning method is employed: the ”supervised” phase quickly incorporates prior knowledge to accelerate learning, while the ”reinforced” phase refines the policy network to overcome the limitations of relying solely on prior knowledge. Finally, experiments were conducted using a mapless navigation task for mobile robots to verify the effectiveness and practical feasibility of the method.

{"title":"Knowledge guided deep deterministic policy gradient","authors":"Peng Qin, Tao Zhao","doi":"10.1016/j.knosys.2025.113087","DOIUrl":"10.1016/j.knosys.2025.113087","url":null,"abstract":"<div><div>Deep deterministic policy gradient (DDPG) exhibits excellent handling capabilities for complex regulation and control problems with continuous state and action spaces. However, its trial-and-error interaction and learning from scratch require extensive exploration by the agent, leading to low learning efficiency and even non-convergence in sparse reward environments. To fully utilize knowledge during the learning process to improve efficiency and performance, this paper draws inspiration from human learning methods and proposes a semantic knowledge-guided DDPG (KGDDPG) approach. In terms of knowledge representation, considering the fuzziness and precision of semantic knowledge, a knowledge system based on a rule framework combining precise propositions and fuzzy propositions is constructed. In terms of knowledge integration, to reduce the randomness of exploration, a knowledge-guided action strategy based on stacked generalization is proposed. Furthermore, a supervised-then-reinforced learning method is employed: the ”supervised” phase quickly incorporates prior knowledge to accelerate learning, while the ”reinforced” phase refines the policy network to overcome the limitations of relying solely on prior knowledge. Finally, experiments were conducted using a mapless navigation task for mobile robots to verify the effectiveness and practical feasibility of the method.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"311 ","pages":"Article 113087"},"PeriodicalIF":7.2,"publicationDate":"2025-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143340035","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}

引用次数: 0

Object tracking using optimized Dual interactive Wasserstein generative adversarial network from surveillance video

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-04 DOI: 10.1016/j.knosys.2025.113084

Karthik Srinivasan

Object tracking in videos is crucial for applications such as video analytics, video surveillance, and intelligent transportation systems. Despite important advancements, challenges like occlusions, background noise, variable object counts, and object appearance similarity still hinder effective tracking. To overcome these complications, Object Tracking using Optimized Dual Interactive Wasserstein Generative Adversarial Network from Surveillance Video (OTSV-DWGAN-GPCOA) is proposed. The input data is collected from the Moving Objects Video Clips Dataset. During the Pre-Processing Phase, noise removal and background subtraction are performed using Anisotropic Diffusion Kuwahara Filtering (ADKF), transforming the surveillance video clips into unique frames for analysis. In the Moving Object Detection Phase, Residual Exemplars Local Binary Pattern (RELBP) is utilized to extract morphological features such as size, texture, color, intensity, shape, and contrast. Additionally, Adaptive Density-Based Spatial Clustering (ADSC) is employed to detect moving objects. In the Moving Object Tracking Phase, the Giza Pyramids Construction Optimization Algorithm (GPCOA) optimizes the parameters of the DWGAN to improve tracking accuracy. Once objects are successfully tracked, the output represents the tracking steps. In the final phase, Moving Object Prediction utilizes the Minkowski Distance Metric to predict the position of tracked objects in each frame. The OTSV-DWGAN-GPCOA method is implemented in Python and assessed using performance metrics. The method achieves 20.11 %, 24.16 % and 22.23 % higher accuracy, 22.45 %, 19.34 % and 24.22 % higher Tracking rate analyzed with existing techniques such asMOD-YOLOv2-SV,ODL-CNN-MRCED, and AR-SSN-VSSC respectively.

{"title":"Object tracking using optimized Dual interactive Wasserstein generative adversarial network from surveillance video","authors":"Karthik Srinivasan","doi":"10.1016/j.knosys.2025.113084","DOIUrl":"10.1016/j.knosys.2025.113084","url":null,"abstract":"<div><div>Object tracking in videos is crucial for applications such as video analytics, video surveillance, and intelligent transportation systems. Despite important advancements, challenges like occlusions, background noise, variable object counts, and object appearance similarity still hinder effective tracking. To overcome these complications, Object Tracking using Optimized Dual Interactive Wasserstein Generative Adversarial Network from Surveillance Video (OTSV-DWGAN-GPCOA) is proposed<strong>.</strong> The input data is collected from the Moving Objects Video Clips Dataset. During the Pre-Processing Phase<strong>,</strong> noise removal and background subtraction are performed using Anisotropic Diffusion Kuwahara Filtering (ADKF)<strong>,</strong> transforming the surveillance video clips into unique frames for analysis. In the Moving Object Detection Phase<strong>,</strong> Residual Exemplars Local Binary Pattern (RELBP) is utilized to extract morphological features such as size, texture, color, intensity, shape, and contrast. Additionally, Adaptive Density-Based Spatial Clustering (ADSC) is employed to detect moving objects. In the Moving Object Tracking Phase<strong>,</strong> the Giza Pyramids Construction Optimization Algorithm (GPCOA) optimizes the parameters of the DWGAN to improve tracking accuracy. Once objects are successfully tracked, the output represents the tracking steps. In the final phase, Moving Object Prediction utilizes the Minkowski Distance Metric to predict the position of tracked objects in each frame. The OTSV-DWGAN-GPCOA method is implemented in Python and assessed using performance metrics. The method achieves 20.11 %, 24.16 % and 22.23 % higher accuracy, 22.45 %, 19.34 % and 24.22 % higher Tracking rate analyzed with existing techniques such asMOD-YOLOv2-SV,ODL-CNN-MRCED, and AR-SSN-VSSC respectively<strong>.</strong></div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"311 ","pages":"Article 113084"},"PeriodicalIF":7.2,"publicationDate":"2025-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143402664","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}

引用次数: 0

Cross-domain recommendation via knowledge distillation

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-04 DOI: 10.1016/j.knosys.2025.113112

Xiuze Li , Zhenhua Huang , Zhengyang Wu , Changdong Wang , Yunwen Chen

Recommendation systems frequently suffer from data sparsity, resulting in less-than-ideal recommendations. A prominent solution to this problem is Cross-Domain Recommendation (CDR), which employs data from various domains to mitigate data sparsity and cold-start issues. Nevertheless, current mainstream methods, like feature mapping and co-training exploring domain relationships, overlook latent user–user and user–item similarities in the shared user–item interaction graph. Spurred by these deficiencies, this paper introduces KDCDR, a novel cross-domain recommendation framework that relies on knowledge distillation to utilize the data from the graph. KDCDR aims to improve the recommendation performance in both domains by efficiently utilizing information from the shared interaction graph. Furthermore, we enhance the effectiveness of user and item representations by exploring the relationships between user–user similarity and item–item similarity, as well as user–item interactions. The developed scheme utilizes the inner-domain graph as a teacher and the cross-domain graph as a student, where the student learns by distilling knowledge from the two teachers after undergoing a high-temperature distillation process. Furthermore, we introduce dynamic weight that regulates the learning process to prevent the student network from overly favoring learning from one domain and focusing on learning knowledge that the teachers have taught incorrectly. Through extensive experiments on four real-world datasets, KDCDR demonstrates significant improvements over state-of-the-art methods, proving the effectiveness of KDCDR in addressing data sparsity issues and enhancing cross-domain recommendation performance. Our code and data are available at https://github.com/pandas-bondage/KDCDR.

{"title":"Cross-domain recommendation via knowledge distillation","authors":"Xiuze Li , Zhenhua Huang , Zhengyang Wu , Changdong Wang , Yunwen Chen","doi":"10.1016/j.knosys.2025.113112","DOIUrl":"10.1016/j.knosys.2025.113112","url":null,"abstract":"<div><div>Recommendation systems frequently suffer from data sparsity, resulting in less-than-ideal recommendations. A prominent solution to this problem is Cross-Domain Recommendation (CDR), which employs data from various domains to mitigate data sparsity and cold-start issues. Nevertheless, current mainstream methods, like feature mapping and co-training exploring domain relationships, overlook latent user–user and user–item similarities in the shared user–item interaction graph. Spurred by these deficiencies, this paper introduces KDCDR, a novel cross-domain recommendation framework that relies on knowledge distillation to utilize the data from the graph. KDCDR aims to improve the recommendation performance in both domains by efficiently utilizing information from the shared interaction graph. Furthermore, we enhance the effectiveness of user and item representations by exploring the relationships between user–user similarity and item–item similarity, as well as user–item interactions. The developed scheme utilizes the inner-domain graph as a teacher and the cross-domain graph as a student, where the student learns by distilling knowledge from the two teachers after undergoing a high-temperature distillation process. Furthermore, we introduce dynamic weight that regulates the learning process to prevent the student network from overly favoring learning from one domain and focusing on learning knowledge that the teachers have taught incorrectly. Through extensive experiments on four real-world datasets, KDCDR demonstrates significant improvements over state-of-the-art methods, proving the effectiveness of KDCDR in addressing data sparsity issues and enhancing cross-domain recommendation performance. Our code and data are available at <span><span>https://github.com/pandas-bondage/KDCDR</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"311 ","pages":"Article 113112"},"PeriodicalIF":7.2,"publicationDate":"2025-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143340029","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}

引用次数: 0

TML-DA: Transfer metric learning based distribution alignment framework for domain class imbalanced classification

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-03 DOI: 10.1016/j.knosys.2025.113086

Mi Yan , Na Jiang , Ning Li

Nonrandom and biased sampling can lead to class imbalance and distribution mismatch issues between domains. Most existing methods sequentially address these issues by combining resampling strategies with transfer learning. However, these approaches typically focus only on the class imbalance within a single domain and overlook the imbalance across domains. To tackle both domain class imbalance and distribution mismatch problems, this paper proposes a transfer metric learning-based distribution alignment (TML-DA) framework, designed for homogeneous and transductive transfer learning. First, the importance-based transfer metric learning module constructs a transfer metric network with an importance parameter, which learns domain-invariant feature representations of source and target data under the domain class imbalance by increasing its within-class coherence and between-class difference. Then, the target domain label prediction module predicts more accurate labels for the unlabeled target data by leveraging inter-domain distance similarity, offering an improvement over traditional probabilistic prediction methods. Finally, the domain distribution alignment module minimizes both marginal and conditional distribution discrepancies while maximizing within-class coherence and between-class difference. This ensures that the learned transfer metric network generalizes more effectively from the source to the target domain. The proposed TML-DA has been evaluated on the long-tailed USPS+MNIST and Office+Caltech public datasets, delivering superior performance and generalization ability in addressing the domain class imbalance classification of the unlabeled target domain data.

{"title":"TML-DA: Transfer metric learning based distribution alignment framework for domain class imbalanced classification","authors":"Mi Yan , Na Jiang , Ning Li","doi":"10.1016/j.knosys.2025.113086","DOIUrl":"10.1016/j.knosys.2025.113086","url":null,"abstract":"<div><div>Nonrandom and biased sampling can lead to class imbalance and distribution mismatch issues between domains. Most existing methods sequentially address these issues by combining resampling strategies with transfer learning. However, these approaches typically focus only on the class imbalance within a single domain and overlook the imbalance across domains. To tackle both domain class imbalance and distribution mismatch problems, this paper proposes a transfer metric learning-based distribution alignment (TML-DA) framework, designed for homogeneous and transductive transfer learning. First, the importance-based transfer metric learning module constructs a transfer metric network with an importance parameter, which learns domain-invariant feature representations of source and target data under the domain class imbalance by increasing its within-class coherence and between-class difference. Then, the target domain label prediction module predicts more accurate labels for the unlabeled target data by leveraging inter-domain distance similarity, offering an improvement over traditional probabilistic prediction methods. Finally, the domain distribution alignment module minimizes both marginal and conditional distribution discrepancies while maximizing within-class coherence and between-class difference. This ensures that the learned transfer metric network generalizes more effectively from the source to the target domain. The proposed TML-DA has been evaluated on the long-tailed USPS+MNIST and Office+Caltech public datasets, delivering superior performance and generalization ability in addressing the domain class imbalance classification of the unlabeled target domain data.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"311 ","pages":"Article 113086"},"PeriodicalIF":7.2,"publicationDate":"2025-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143372779","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}

引用次数: 0

首页上一页

下一页尾页

类型

全部化学•材料生命科学医学物理工程技术环境•农林材料科学地球科学法学管理学化学环境科学与生态学计算机科学教育学经济学农林科学人文科学生物学数学物理与天体物理心理学综合性期刊其他工业工程理学历史学农学文学信息工程

数据库

全部 ACS Publications Elsevier ieeexplore Springer The Royal Society of Chemistry Wiley

期刊

Knowledge-Based Systems

全部 Acc. Chem. Res. ACS Applied Bio Materials ACS Appl. Electron. Mater. ACS Appl. Energy Mater. ACS Appl. Mater. Interfaces ACS Appl. Nano Mater. ACS Appl. Polym. Mater. ACS BIOMATER-SCI ENG ACS Catal. ACS Cent. Sci. ACS Chem. Biol. ACS Chemical Health & Safety ACS Chem. Neurosci. ACS Comb. Sci. ACS Earth Space Chem. ACS Energy Lett. ACS Infect. Dis. ACS Macro Lett. ACS Mater. Lett. ACS Med. Chem. Lett. ACS Nano ACS Omega ACS Photonics ACS Sens. ACS Sustainable Chem. Eng. ACS Synth. Biol. Anal. Chem. BIOCHEMISTRY-US Bioconjugate Chem. BIOMACROMOLECULES Chem. Res. Toxicol. Chem. Rev. Chem. Mater. CRYST GROWTH DES ENERG FUEL Environ. Sci. Technol. Environ. Sci. Technol. Lett. Eur. J. Inorg. Chem. IND ENG CHEM RES Inorg. Chem. J. Agric. Food. Chem. J. Chem. Eng. Data J. Chem. Educ. J. Chem. Inf. Model. J. Chem. Theory Comput. J. Med. Chem. J. Nat. Prod. J PROTEOME RES J. Am. Chem. Soc. LANGMUIR MACROMOLECULES Mol. Pharmaceutics Nano Lett. Org. Lett. ORG PROCESS RES DEV ORGANOMETALLICS J. Org. Chem. J. Phys. Chem. J. Phys. Chem. A J. Phys. Chem. B J. Phys. Chem. C J. Phys. Chem. Lett. Analyst Anal. Methods Biomater. Sci. Catal. Sci. Technol. Chem. Commun. Chem. Soc. Rev. CHEM EDUC RES PRACT CRYSTENGCOMM Dalton Trans. Energy Environ. Sci. ENVIRON SCI-NANO ENVIRON SCI-PROC IMP ENVIRON SCI-WAT RES Faraday Discuss. Food Funct. Green Chem. Inorg. Chem. Front. Integr. Biol. J. Anal. At. Spectrom. J. Mater. Chem. A J. Mater. Chem. B J. Mater. Chem. C Lab Chip Mater. Chem. Front. Mater. Horiz. MEDCHEMCOMM Metallomics Mol. Biosyst. Mol. Syst. Des. Eng. Nanoscale Nanoscale Horiz. Nat. Prod. Rep. New J. Chem. Org. Biomol. Chem. Org. Chem. Front. PHOTOCH PHOTOBIO SCI PCCP Polym. Chem.

﹀