Pub Date : 2025-02-22DOI: 10.1016/j.asoc.2025.112894
Beibei Zhu , Ruijie Tian , Xiaosong Yuan , Ridong Han , Yan Yang , Bo Fu
Knowledge graph entity alignment seeks to match equivalent entities across different graphs, a critical task for enabling cross-lingual knowledge fusion. Mainstream methods use representation learning for entity alignment based on vector distances, but struggle with complex relational semantics and underutilize fine-grained attribute information crucial for alignment. To overcome the above problems, this paper proposes a cross-lingual entity alignment model based on complex relationships and fine-grained attributes (CEARA). The proposed model effectively handles relational semantics by distinguishing their varying impacts on entity embeddings and extracting detailed attribute information to enhance alignment accuracy. Additionally, it integrates entity name string similarity to complement missing or noisy relational and attribute data, further improving alignment reliability. To mitigate alignment conflicts, the model employs a global alignment strategy. Experimental results on three cross-lingual datasets demonstrate that CEARA not only outperforms representative baseline models but also achieves Hits@1 scores exceeding 95% across all datasets, highlighting its effectiveness and robustness for cross-lingual alignment. This paper contributes to the advancement of cross-lingual knowledge discovery and application.
{"title":"Cross-lingual entity alignment based on complex relationships and fine-grained attributes","authors":"Beibei Zhu , Ruijie Tian , Xiaosong Yuan , Ridong Han , Yan Yang , Bo Fu","doi":"10.1016/j.asoc.2025.112894","DOIUrl":"10.1016/j.asoc.2025.112894","url":null,"abstract":"<div><div>Knowledge graph entity alignment seeks to match equivalent entities across different graphs, a critical task for enabling cross-lingual knowledge fusion. Mainstream methods use representation learning for entity alignment based on vector distances, but struggle with complex relational semantics and underutilize fine-grained attribute information crucial for alignment. To overcome the above problems, this paper proposes a <strong>c</strong>ross-lingual <strong>e</strong>ntity <strong>a</strong>lignment model based on complex <strong>r</strong>elationships and fine-grained <strong>a</strong>ttributes (CEARA). The proposed model effectively handles relational semantics by distinguishing their varying impacts on entity embeddings and extracting detailed attribute information to enhance alignment accuracy. Additionally, it integrates entity name string similarity to complement missing or noisy relational and attribute data, further improving alignment reliability. To mitigate alignment conflicts, the model employs a global alignment strategy. Experimental results on three cross-lingual datasets demonstrate that CEARA not only outperforms representative baseline models but also achieves Hits@1 scores exceeding 95% across all datasets, highlighting its effectiveness and robustness for cross-lingual alignment. This paper contributes to the advancement of cross-lingual knowledge discovery and application.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112894"},"PeriodicalIF":7.2,"publicationDate":"2025-02-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143474146","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}
This study explores the use of Pythagorean fuzzy soft sets (PFSSs) in group decision-making to handle the uncertainty and ambiguity in information data. Aggregation operators take an important role in separating problems, from the perception of two prospect circulations and performing perceptivity between them. In this article, we generalize the notion of interval-valued intuitionistic fuzzy soft sets (IVIFSSs) into interval-valued PFSSs (IVPFSSs) to enhance decision-making processes under conditions of uncertainty and ambiguity. Compared to the current interval-valued Pythagorean fuzzy set, the IVPFSSs handle uncertain and ambiguous information with efficiency. To support this framework, new operational laws for IVPFSSs are developed. Furthermore, two innovative aggregation operators i.e. the interval-valued Pythagorean fuzzy soft set weighted average (IVPFSSWA) and the interval-valued Pythagorean fuzzy soft set weighted geometric (IVPFSSWG)—have been constructed based on the operational laws with their fundamental properties. These operators facilitate the aggregation of fuzzy data effectively and reliably. To address this, a resilient multi-criteria decision-making (MCDM) technique is designed using the proposed aggregation operators, specifically for material selection in product design and manufacturing. A real-world application illustrates the practicality of the method, focusing on material selection. The results highlight the model’s effectiveness and reliability in handling fuzzy data based on IVPFSSs. Finally, the results are compared with some existing operators to check the reliability of the proposed aggregation operators. In addition, a comprehensive comparison analysis is carried out to establish the enhanced performance, feasibility and robustness of the obtained results. However, there are number of shortcomings to the suggested operators, such as their high computational complexity, sensitivity to weight assignments, complexity in interpretability and difficulties in managing uncertainty.
{"title":"An innovative aggregation operator for enhanced decision-making: A study on interval-valued Pythagorean fuzzy soft sets in material selection","authors":"Diptirekha Sahoo , Prashanta Kumar Parida , Sandhya Priya Baral , Bibudhendu Pati","doi":"10.1016/j.asoc.2025.112888","DOIUrl":"10.1016/j.asoc.2025.112888","url":null,"abstract":"<div><div>This study explores the use of Pythagorean fuzzy soft sets (PFSSs) in group decision-making to handle the uncertainty and ambiguity in information data. Aggregation operators take an important role in separating problems, from the perception of two prospect circulations and performing perceptivity between them. In this article, we generalize the notion of interval-valued intuitionistic fuzzy soft sets (IVIFSSs) into interval-valued PFSSs (IVPFSSs) to enhance decision-making processes under conditions of uncertainty and ambiguity. Compared to the current interval-valued Pythagorean fuzzy set, the IVPFSSs handle uncertain and ambiguous information with efficiency. To support this framework, new operational laws for IVPFSSs are developed. Furthermore, two innovative aggregation operators i.e. the interval-valued Pythagorean fuzzy soft set weighted average (IVPFSSWA) and the interval-valued Pythagorean fuzzy soft set weighted geometric (IVPFSSWG)—have been constructed based on the operational laws with their fundamental properties. These operators facilitate the aggregation of fuzzy data effectively and reliably. To address this, a resilient multi-criteria decision-making (MCDM) technique is designed using the proposed aggregation operators, specifically for material selection in product design and manufacturing. A real-world application illustrates the practicality of the method, focusing on material selection. The results highlight the model’s effectiveness and reliability in handling fuzzy data based on IVPFSSs. Finally, the results are compared with some existing operators to check the reliability of the proposed aggregation operators. In addition, a comprehensive comparison analysis is carried out to establish the enhanced performance, feasibility and robustness of the obtained results. However, there are number of shortcomings to the suggested operators, such as their high computational complexity, sensitivity to weight assignments, complexity in interpretability and difficulties in managing uncertainty.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112888"},"PeriodicalIF":7.2,"publicationDate":"2025-02-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143474053","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}
Pub Date : 2025-02-20DOI: 10.1016/j.asoc.2025.112873
Pengguo Yan , Ye Tian , Jiesheng Wang , Yu Liu
Constrained multi-objective optimization problems (CMOPs) present significant challenges due to the simultaneous consideration of objectives and constraints, which becomes particularly arduous when the feasible regions are exceedingly complex. Most of the existing algorithms fail to obtain high-quality solutions for the CMOPs with complex feasible regions. To address this issue, this paper proposes a Constraint Priority Decision framework applied to multi-stage evolutionary algorithms, which incorporates constraints sequentially throughout the solution process to facilitate the retention of optimal diversity and feasibility within the population. Specifically, the proposed framework employs a traditional multi-objective evolutionary algorithm as the optimizer and decomposes various constraints of the CMOP. These constraints are introduced independently into the optimizer, generating an index value for each respective constraint. Following this, a judgment matrix is constructed based on these indices to grade the constraints, thus facilitating the establishment of a priority sequence for multiple constraints. Furthermore, a two-stage strategy is implemented in this study. After incorporating all constraints into the algorithm, the -constrained method is utilized to impose constraints on the entire problem to increase the genetic diversity of the population while maintaining the feasibility of the population. The experimental results derived from four popular benchmark suites and six real-world applications indicate that the proposed framework surpassed multiple state-of-the-art constrained multi-objective evolutionary algorithms in addressing CMOPs with complex feasible regions.
{"title":"A Constraint Priority Decision framework for constrained multi-objective optimization with complex feasible regions","authors":"Pengguo Yan , Ye Tian , Jiesheng Wang , Yu Liu","doi":"10.1016/j.asoc.2025.112873","DOIUrl":"10.1016/j.asoc.2025.112873","url":null,"abstract":"<div><div>Constrained multi-objective optimization problems (CMOPs) present significant challenges due to the simultaneous consideration of objectives and constraints, which becomes particularly arduous when the feasible regions are exceedingly complex. Most of the existing algorithms fail to obtain high-quality solutions for the CMOPs with complex feasible regions. To address this issue, this paper proposes a Constraint Priority Decision framework applied to multi-stage evolutionary algorithms, which incorporates constraints sequentially throughout the solution process to facilitate the retention of optimal diversity and feasibility within the population. Specifically, the proposed framework employs a traditional multi-objective evolutionary algorithm as the optimizer and decomposes various constraints of the CMOP. These constraints are introduced independently into the optimizer, generating an index value for each respective constraint. Following this, a judgment matrix is constructed based on these indices to grade the constraints, thus facilitating the establishment of a priority sequence for multiple constraints. Furthermore, a two-stage strategy is implemented in this study. After incorporating all constraints into the algorithm, the <span><math><mi>ϵ</mi></math></span>-constrained method is utilized to impose constraints on the entire problem to increase the genetic diversity of the population while maintaining the feasibility of the population. The experimental results derived from four popular benchmark suites and six real-world applications indicate that the proposed framework surpassed multiple state-of-the-art constrained multi-objective evolutionary algorithms in addressing CMOPs with complex feasible regions.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112873"},"PeriodicalIF":7.2,"publicationDate":"2025-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143474145","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}
Pub Date : 2025-02-20DOI: 10.1016/j.asoc.2025.112874
Mingjing Wang , Xiaoping Li , Long Chen , Huiling Chen , Chi Chen , Minzhe Liu
As the number of decision variables increases, the curse of dimensionality becomes a significant challenge in many practical multi-objective optimization problems. This issue is further exacerbated in large-scale many-objective optimization problems (MaOPs), where the growing number of optimization objectives makes it increasingly difficult for evolutionary algorithms to find optimal solutions. In this study, we propose a deep Gaussian mixture model algorithm tailored for large-scale MaOPs. The novelty of this approach lies in its hierarchical detection of interactions and redundancies among decision variables, enabling a more effective grouping of variables. Specifically, a Gaussian mixture model-based framework is used to model the problem, allowing for the preliminary grouping of decision variables. The proposed Grouping Decision Variables using the Gaussian Mixture Model (GDVG) algorithm categorizes variables into two types: convergence-related and diversity-related variables. Additionally, a Linkage Identification Measurement with Chaos (LIMC) method is introduced for grouping convergence-related variables based on their interactions. For diversity-related variables, we present a Trivial Variable Detection Scheme (TVDS) to identify and group variables that contribute to diversity. The experimental results demonstrate that the proposed method outperforms other competitive algorithms on most benchmark test cases, particularly showcasing its effectiveness in large-scale MaOPs.
{"title":"A deep-based Gaussian mixture model algorithm for large-scale many objective optimization","authors":"Mingjing Wang , Xiaoping Li , Long Chen , Huiling Chen , Chi Chen , Minzhe Liu","doi":"10.1016/j.asoc.2025.112874","DOIUrl":"10.1016/j.asoc.2025.112874","url":null,"abstract":"<div><div>As the number of decision variables increases, the curse of dimensionality becomes a significant challenge in many practical multi-objective optimization problems. This issue is further exacerbated in large-scale many-objective optimization problems (MaOPs), where the growing number of optimization objectives makes it increasingly difficult for evolutionary algorithms to find optimal solutions. In this study, we propose a deep Gaussian mixture model algorithm tailored for large-scale MaOPs. The novelty of this approach lies in its hierarchical detection of interactions and redundancies among decision variables, enabling a more effective grouping of variables. Specifically, a Gaussian mixture model-based framework is used to model the problem, allowing for the preliminary grouping of decision variables. The proposed Grouping Decision Variables using the Gaussian Mixture Model (GDVG) algorithm categorizes variables into two types: convergence-related and diversity-related variables. Additionally, a Linkage Identification Measurement with Chaos (LIMC) method is introduced for grouping convergence-related variables based on their interactions. For diversity-related variables, we present a Trivial Variable Detection Scheme (TVDS) to identify and group variables that contribute to diversity. The experimental results demonstrate that the proposed method outperforms other competitive algorithms on most benchmark test cases, particularly showcasing its effectiveness in large-scale MaOPs.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112874"},"PeriodicalIF":7.2,"publicationDate":"2025-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464775","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}
Pub Date : 2025-02-20DOI: 10.1016/j.asoc.2025.112875
Haejin Jang , Dojin Kim , Hyeonseo Kim , Lee-Chae Jang
This paper presents a novel application of the triangular fuzzy number (TFN)-generalized Choquet integral to Hofstede’s cultural dimensions data from 50 countries, introducing a robust methodology for analyzing cultural diversity and its economic implications. A 5-level standard evaluation is proposed as an integrated value indicator, enabling a comprehensive interpretation of Hofstede’s cultural dimensions and their relationships with key economic indicators, including GDP, unemployment rate, and population growth rate. To operationalize this evaluation, an algorithm is developed to quantify cultural dimensions into standardized linguistic symbols based on the 5-level evaluation. This algorithm integrates the TFN-generalized Choquet integral with fuzzy measures and a utility function, facilitating the classification of countries into tendency levels for comparative analysis. The algorithm enhances interpretability by mapping numerical data to intuitive categories, making the findings actionable for policymakers. Through a comparative analysis, the study identifies role model countries exhibiting exemplary development patterns and explores correlations between cultural tendency levels and economic performance. The findings validate three hypotheses linking cultural dimensions to economic outcomes and provide actionable recommendations to enhance economic growth, employment, and population dynamics. This framework equips policymakers with valuable tools for leveraging cultural awareness to drive sustainable economic development.
{"title":"Applications of Hofstede’s cultural dimensions in 50 countries using TFN-generalized Choquet integrals","authors":"Haejin Jang , Dojin Kim , Hyeonseo Kim , Lee-Chae Jang","doi":"10.1016/j.asoc.2025.112875","DOIUrl":"10.1016/j.asoc.2025.112875","url":null,"abstract":"<div><div>This paper presents a novel application of the triangular fuzzy number (TFN)-generalized Choquet integral to Hofstede’s cultural dimensions data from 50 countries, introducing a robust methodology for analyzing cultural diversity and its economic implications. A 5-level standard evaluation is proposed as an integrated value indicator, enabling a comprehensive interpretation of Hofstede’s cultural dimensions and their relationships with key economic indicators, including GDP, unemployment rate, and population growth rate. To operationalize this evaluation, an algorithm is developed to quantify cultural dimensions into standardized linguistic symbols based on the 5-level evaluation. This algorithm integrates the TFN-generalized Choquet integral with fuzzy measures and a utility function, facilitating the classification of countries into tendency levels for comparative analysis. The algorithm enhances interpretability by mapping numerical data to intuitive categories, making the findings actionable for policymakers. Through a comparative analysis, the study identifies role model countries exhibiting exemplary development patterns and explores correlations between cultural tendency levels and economic performance. The findings validate three hypotheses linking cultural dimensions to economic outcomes and provide actionable recommendations to enhance economic growth, employment, and population dynamics. This framework equips policymakers with valuable tools for leveraging cultural awareness to drive sustainable economic development.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112875"},"PeriodicalIF":7.2,"publicationDate":"2025-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143471606","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}
Pub Date : 2025-02-19DOI: 10.1016/j.asoc.2025.112897
Wei Zhou , Wenqiang Zhu , Jin Chen , Zeshui Xu
Discretization is a widely used technique for organizing and simplifying continuous-valued data in classification problems, facilitating subsequent analytical applications. However, the lack of an effective approach for addressing soft boundary classification can complicate the interpretation and accuracy of results. This paper focuses on discretization methods within soft boundary classification environments, aiming to improve classification outcomes and enhance interpretability for specific applications. Building on rule-based discretization, we propose a novel approach to enhance the handling of soft boundary classification. A detailed illustrative example is provided to demonstrate the effectiveness of the proposed Cross-Interval Recursion (CIR) method and Heuristic Cross-Interval Recursion (HCIR) algorithm. Our results show that the CIR-based rule discretization method and its evaluation mechanism effectively mitigate noise interference from class boundary points, improving interpretability and promoting greater generalization in soft boundary classification. The performance of our algorithm outperforms existing methods, including EqQua-CIR, EqVal-CIR, and other rule-based discretization techniques, particularly in terms of classification accuracy when dealing with boundary points at high granularity. When compared to classic classifiers and other rule-based discretization approaches, our method demonstrates that rule-based classifiers are more effective than direct approaches in handling soft boundary issues. Furthermore, the alignment between the classifier and the sample data plays a critical role in determining classification performance. Our approach offers significant potential as a breakthrough in addressing soft boundary classification of continuous-valued attributes, leveraging interval reconstruction, and enhancing classification robustness.
{"title":"The cross-interval reconstruction and heuristic calculation to deal with the continuous-valued attribute in the learning process","authors":"Wei Zhou , Wenqiang Zhu , Jin Chen , Zeshui Xu","doi":"10.1016/j.asoc.2025.112897","DOIUrl":"10.1016/j.asoc.2025.112897","url":null,"abstract":"<div><div>Discretization is a widely used technique for organizing and simplifying continuous-valued data in classification problems, facilitating subsequent analytical applications. However, the lack of an effective approach for addressing soft boundary classification can complicate the interpretation and accuracy of results. This paper focuses on discretization methods within soft boundary classification environments, aiming to improve classification outcomes and enhance interpretability for specific applications. Building on rule-based discretization, we propose a novel approach to enhance the handling of soft boundary classification. A detailed illustrative example is provided to demonstrate the effectiveness of the proposed Cross-Interval Recursion (CIR) method and Heuristic Cross-Interval Recursion (HCIR) algorithm. Our results show that the CIR-based rule discretization method and its evaluation mechanism effectively mitigate noise interference from class boundary points, improving interpretability and promoting greater generalization in soft boundary classification. The performance of our algorithm outperforms existing methods, including EqQua-CIR, EqVal-CIR, and other rule-based discretization techniques, particularly in terms of classification accuracy when dealing with boundary points at high granularity. When compared to classic classifiers and other rule-based discretization approaches, our method demonstrates that rule-based classifiers are more effective than direct approaches in handling soft boundary issues. Furthermore, the alignment between the classifier and the sample data plays a critical role in determining classification performance. Our approach offers significant potential as a breakthrough in addressing soft boundary classification of continuous-valued attributes, leveraging interval reconstruction, and enhancing classification robustness.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112897"},"PeriodicalIF":7.2,"publicationDate":"2025-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143455105","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}
Pub Date : 2025-02-18DOI: 10.1016/j.asoc.2025.112871
Wenhan Liu , Shurong Pan , Sheng Chang , Qijun Huang , Nan Jiang
In recent years, the development of deep learning has shown potential in the automatic analysis of electrocardiogram (ECG), aiding cardiologists in detecting cardiovascular diseases (CVDs). Generally, deep learning models depend on numerous labeled ECGs to train, but manual labeling of ECGs is costly as it requires considerable time and expertise. Self-supervised learning (SSL) can solve this problem by pretraining deep learning models with unlabeled ECGs, mitigating their reliance on labeled ECGs. This work proposes lead correlation and decorrelation (LCD) for effective and efficient SSL of ECGs. Concretely, LCD combines intra-lead correlation, inter-lead correlation, intra-lead and inter-lead decorrelation in pretraining. These mechanisms utilize multilead ECG characteristics: intra-lead invariance, inter-lead invariance, inter-lead variance, and intra-lead redundancy. After pretraining, LCD can provide a generic encoder for feature extraction of any ECG lead in a classification task. Benefitting from the effective pretraining mechanism, models with the encoders pretrained by LCD outperform most of the baselines. Compared with the best baseline, they achieve better/comparable classification performances in the same tasks with less pretraining time. Furthermore, LCD helps the models focus on critical features when training with insufficient labeled ECGs, reducing the reliance on labeled ECGs by 46. All the results demonstrate that LCD is an effective and efficient method, boosting a broader application of deep learning to automatic ECG analysis. The code is available at https://github.com/Aiwiscal/ECG_SSL_LCD.
{"title":"Self-supervised learning for Electrocardiogram classification using Lead Correlation and Decorrelation","authors":"Wenhan Liu , Shurong Pan , Sheng Chang , Qijun Huang , Nan Jiang","doi":"10.1016/j.asoc.2025.112871","DOIUrl":"10.1016/j.asoc.2025.112871","url":null,"abstract":"<div><div>In recent years, the development of deep learning has shown potential in the automatic analysis of electrocardiogram (ECG), aiding cardiologists in detecting cardiovascular diseases (CVDs). Generally, deep learning models depend on numerous labeled ECGs to train, but manual labeling of ECGs is costly as it requires considerable time and expertise. Self-supervised learning (SSL) can solve this problem by pretraining deep learning models with unlabeled ECGs, mitigating their reliance on labeled ECGs. This work proposes lead correlation and decorrelation (LCD) for effective and efficient SSL of ECGs. Concretely, LCD combines intra-lead correlation, inter-lead correlation, intra-lead and inter-lead decorrelation in pretraining. These mechanisms utilize multilead ECG characteristics: intra-lead invariance, inter-lead invariance, inter-lead variance, and intra-lead redundancy. After pretraining, LCD can provide a generic encoder for feature extraction of any ECG lead in a classification task. Benefitting from the effective pretraining mechanism, models with the encoders pretrained by LCD outperform most of the baselines. Compared with the best baseline, they achieve better/comparable classification performances in the same tasks with less pretraining time. Furthermore, LCD helps the models focus on critical features when training with insufficient labeled ECGs, reducing the reliance on labeled ECGs by 4<span><math><mo>∼</mo></math></span>6<span><math><mo>×</mo></math></span>. All the results demonstrate that LCD is an effective and efficient method, boosting a broader application of deep learning to automatic ECG analysis. The code is available at <span><span>https://github.com/Aiwiscal/ECG_SSL_LCD</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112871"},"PeriodicalIF":7.2,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143455104","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}
Pub Date : 2025-02-18DOI: 10.1016/j.asoc.2025.112889
Liying Wang, Weiguo Zhao
Intelligent classification methods based on deep learning (DL) have become widely adopted for bearing fault diagnosis (BFD). However, it is acknowledged that relying on single feature extraction methods may not yield comprehensive representations of the information features. Additionally, DL-based approaches for extracting features from vibration signals typically utilize either one-dimensional (1D) or two-dimensional (2D) networks, which can restrict the network's ability to extract features effectively. In this paper, a time series data representation method called the relative angle matrix (RAM) method is firstly proposed. This method converts 1D time series into 2D images by calculating the angle differences between multiple vectors and a central vector, thereby extracting the hidden spatial features present in the original data. Then, this paper introduces an ensemble deep learning network called 1D2D-EDL, which integrates 1D-based and 2D-based DL mechanisms for feature extraction and classification, leveraging the strengths of each approach. The 1D2D-EDL comprises two channels: the 1D channel combines long short-term memory (LSTM) and multi-head self-attention (MSA) to process raw 1D time series data, facilitating feature extraction in both the time and frequency domains. Meanwhile, the 2D channel employs convolutional neural network (CNN) components to process 2D images for spatial feature extraction, which are derived from the original time series data using the RAM method. Finally, the feature information from these two channels is fused using a feature fusion method. To preliminarily validate the effectiveness of the RAM method, three competitive 2D conversion methods are employed, including Gramian angular difference field (GADF), Gramian angular sum field (GASF), and Markov transition field (MTF). These methods are applied alongside the proposed RAM method within the same CNN network for fault diagnosis testing. The results indicate that the RAM method significantly enhances the diagnostic accuracy of the CNN compared to the other 2D conversion methods. Furthermore, the bearing fault dataset from the University of Ottawa is utilized to validate the performance of the 1D2D-EDL. A comparative analysis with other DL methods using multiple statistical metrics demonstrates the superiority of the 1D2D-EDL. Specifically, when diagnosing faults under four different speed conditions, the 1D2D-EDL attains accuracy rates of 100 %, 99.33 %, 100 %, and 100 %, respectively. This study proposes the incorporation of a novel perspective classifier to enhance DL models for bearing fault diagnosis. The source code of RAM is available at https://ww2.mathworks.cn/matlabcentral/fileexchange/180197-relative-angle-matrix-ram.
{"title":"An ensemble deep learning network based on 2D convolutional neural network and 1D LSTM with self-attention for bearing fault diagnosis","authors":"Liying Wang, Weiguo Zhao","doi":"10.1016/j.asoc.2025.112889","DOIUrl":"10.1016/j.asoc.2025.112889","url":null,"abstract":"<div><div>Intelligent classification methods based on deep learning (DL) have become widely adopted for bearing fault diagnosis (BFD). However, it is acknowledged that relying on single feature extraction methods may not yield comprehensive representations of the information features. Additionally, DL-based approaches for extracting features from vibration signals typically utilize either one-dimensional (1D) or two-dimensional (2D) networks, which can restrict the network's ability to extract features effectively. In this paper, a time series data representation method called the relative angle matrix (RAM) method is firstly proposed. This method converts 1D time series into 2D images by calculating the angle differences between multiple vectors and a central vector, thereby extracting the hidden spatial features present in the original data. Then, this paper introduces an ensemble deep learning network called 1D2D-EDL, which integrates 1D-based and 2D-based DL mechanisms for feature extraction and classification, leveraging the strengths of each approach. The 1D2D-EDL comprises two channels: the 1D channel combines long short-term memory (LSTM) and multi-head self-attention (MSA) to process raw 1D time series data, facilitating feature extraction in both the time and frequency domains. Meanwhile, the 2D channel employs convolutional neural network (CNN) components to process 2D images for spatial feature extraction, which are derived from the original time series data using the RAM method. Finally, the feature information from these two channels is fused using a feature fusion method. To preliminarily validate the effectiveness of the RAM method, three competitive 2D conversion methods are employed, including Gramian angular difference field (GADF), Gramian angular sum field (GASF), and Markov transition field (MTF). These methods are applied alongside the proposed RAM method within the same CNN network for fault diagnosis testing. The results indicate that the RAM method significantly enhances the diagnostic accuracy of the CNN compared to the other 2D conversion methods. Furthermore, the bearing fault dataset from the University of Ottawa is utilized to validate the performance of the 1D2D-EDL. A comparative analysis with other DL methods using multiple statistical metrics demonstrates the superiority of the 1D2D-EDL. Specifically, when diagnosing faults under four different speed conditions, the 1D2D-EDL attains accuracy rates of 100 %, 99.33 %, 100 %, and 100 %, respectively. This study proposes the incorporation of a novel perspective classifier to enhance DL models for bearing fault diagnosis. The source code of RAM is available at <span><span>https://ww2.mathworks.cn/matlabcentral/fileexchange/180197-relative-angle-matrix-ram</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112889"},"PeriodicalIF":7.2,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143474054","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}
Pub Date : 2025-02-18DOI: 10.1016/j.asoc.2025.112890
Saleh Alsulamy
Construction projects in Saudi Arabia often encounter delays, which present significant challenges to project managers and result in financial losses and stakeholder dissatisfaction. Effectively managing these delays is essential for maintaining project timelines and optimizing resource use. This study explores the hypothesis that advanced deep learning algorithms can significantly improve the prediction and management of construction project delays in Saudi Arabia. The research focuses on three algorithms: Generative Adversarial Networks (GAN), Long Short-Term Memory (LSTM), and Multilayer Perceptron (MLP), evaluating their effectiveness across datasets with varying class imbalances. A structured methodology was employed to assess the algorithms based on key performance metrics, including accuracy, precision, sensitivity, specificity, and misclassification errors. GAN, LSTM, and MLP were trained and tested using real-world construction project data, incorporating tools such as k-fold cross-validation for validation. The GAN model achieved the highest accuracy at 91 %, with a misclassification error of 9 %, outperforming both LSTM (accuracy: 88 %, error: 12 %) and MLP (accuracy: 83 %, error: 17 %). GAN also demonstrated superior precision (90 %) and sensitivity (87 %), making it the most reliable algorithm for delay risk assessment. While LSTM was effective, it had slightly lower precision (88 %) but exhibited strong generalization to unseen data. MLP showed the weakest performance, with higher misclassification rates and lower robustness. These findings suggest that deep learning models, particularly GAN, can significantly improve decision-making and delay mitigation in construction projects.
{"title":"Comparative analysis of deep learning algorithms for predicting construction project delays in Saudi Arabia","authors":"Saleh Alsulamy","doi":"10.1016/j.asoc.2025.112890","DOIUrl":"10.1016/j.asoc.2025.112890","url":null,"abstract":"<div><div>Construction projects in Saudi Arabia often encounter delays, which present significant challenges to project managers and result in financial losses and stakeholder dissatisfaction. Effectively managing these delays is essential for maintaining project timelines and optimizing resource use. This study explores the hypothesis that advanced deep learning algorithms can significantly improve the prediction and management of construction project delays in Saudi Arabia. The research focuses on three algorithms: Generative Adversarial Networks (GAN), Long Short-Term Memory (LSTM), and Multilayer Perceptron (MLP), evaluating their effectiveness across datasets with varying class imbalances. A structured methodology was employed to assess the algorithms based on key performance metrics, including accuracy, precision, sensitivity, specificity, and misclassification errors. GAN, LSTM, and MLP were trained and tested using real-world construction project data, incorporating tools such as k-fold cross-validation for validation. The GAN model achieved the highest accuracy at 91 %, with a misclassification error of 9 %, outperforming both LSTM (accuracy: 88 %, error: 12 %) and MLP (accuracy: 83 %, error: 17 %). GAN also demonstrated superior precision (90 %) and sensitivity (87 %), making it the most reliable algorithm for delay risk assessment. While LSTM was effective, it had slightly lower precision (88 %) but exhibited strong generalization to unseen data. MLP showed the weakest performance, with higher misclassification rates and lower robustness. These findings suggest that deep learning models, particularly GAN, can significantly improve decision-making and delay mitigation in construction projects.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112890"},"PeriodicalIF":7.2,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143464776","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}
Pub Date : 2025-02-17DOI: 10.1016/j.asoc.2025.112863
Shuhan Du, Junbo Tong, Wenhui Fan
Mixed Integer Linear Programs (MILPs) are widely used to model various real-world optimization problems, traditionally solved using the branch-and-bound (B&B) algorithm framework. Recent advances in Machine Learning (ML) have inspired enhancements in B&B by enabling data-driven decision-making. Two critical decisions in B&B are node selection and variable selection, which directly influence computational efficiency. While prior studies have applied ML to enhance these decisions, they have predominantly focused on either node selection or variable selection, addressing the decision individually and overlooking the significant interdependence between the two. This paper introduces a novel ML-based approach that integrates both decisions within the B&B framework using a unified neural network architecture. By leveraging a bipartite graph representation of MILPs and employing Graph Neural Networks, the model learns adaptive strategies tailored to different problem types through imitation of expert-designed policies. Experiments on various benchmarks show that the integrated policy adapts better to different problem classes than models targeting individual decisions, delivering strong performance in solving time, search tree size, and optimization dynamics across various configurations. It also surpasses competitive baselines, including the state-of-the-art open-source solver SCIP and a recent reinforcement learning-based approach, demonstrating its potential for broader application in MILP solving.
{"title":"Learning efficient branch-and-bound for solving Mixed Integer Linear Programs","authors":"Shuhan Du, Junbo Tong, Wenhui Fan","doi":"10.1016/j.asoc.2025.112863","DOIUrl":"10.1016/j.asoc.2025.112863","url":null,"abstract":"<div><div>Mixed Integer Linear Programs (MILPs) are widely used to model various real-world optimization problems, traditionally solved using the branch-and-bound (B&B) algorithm framework. Recent advances in Machine Learning (ML) have inspired enhancements in B&B by enabling data-driven decision-making. Two critical decisions in B&B are node selection and variable selection, which directly influence computational efficiency. While prior studies have applied ML to enhance these decisions, they have predominantly focused on either node selection or variable selection, addressing the decision individually and overlooking the significant interdependence between the two. This paper introduces a novel ML-based approach that integrates both decisions within the B&B framework using a unified neural network architecture. By leveraging a bipartite graph representation of MILPs and employing Graph Neural Networks, the model learns adaptive strategies tailored to different problem types through imitation of expert-designed policies. Experiments on various benchmarks show that the integrated policy adapts better to different problem classes than models targeting individual decisions, delivering strong performance in solving time, search tree size, and optimization dynamics across various configurations. It also surpasses competitive baselines, including the state-of-the-art open-source solver SCIP and a recent reinforcement learning-based approach, demonstrating its potential for broader application in MILP solving.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"172 ","pages":"Article 112863"},"PeriodicalIF":7.2,"publicationDate":"2025-02-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143429859","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}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号