Scholarlink

<p>Entity Resolution (ER) is a critical task in data management, where the challenge of imbalanced binary classification frequently arises, leading to significant difficulties in accurately matching records. This study addresses this challenge by introducing a novel approach, the Linear Exponential Loss Function Fuzzy Soft Support Vector Machine (LF-SSVM). The proposed model enhances the traditional Fuzzy Soft SVM by integrating a Linear Exponential (LINEX) loss function, which is specifically designed to handle the complexities associated with imbalanced datasets. The LINEX loss function, recognized for its asymmetric properties, effectively addresses misclassifications by treating them differently based on their characteristics, thereby improving the accuracy of ER in imbalanced scenarios. Additionally, the model is specifically designed to enhance the identification and matching of entities in datasets where certain classes are underrepresented. By utilizing a Primal-Dual optimization method, the model ensures robustness against noisy and imbalanced data while preserving the core structure of the LF-SSVM. Extensive experiments conducted on fifteen real-world ER datasets confirm the effectiveness of our approach, demonstrating substantial improvements over traditional methods in handling imbalanced class distributions in ER tasks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Primal-dual method to solve linear exponential loss function in fuzzy soft SVM: a robust solution to imbalanced data in entity resolution

  • 期刊论文

  • 发表日期：

    2026年5月25日

查看全文
  • Mourad Jabrane,
  • Soufiane Lyaqini,
  • Aissam Hadri,
  • Imad Hafidi

摘要

Entity Resolution (ER) is a critical task in data management, where the challenge of imbalanced binary classification frequently arises, leading to significant difficulties in accurately matching records. This study addresses this challenge by introducing a novel approach, the Linear Exponential Loss Function Fuzzy Soft Support Vector Machine (LF-SSVM). The proposed model enhances the traditional Fuzzy Soft SVM by integrating a Linear Exponential (LINEX) loss function, which is specifically designed to handle the complexities associated with imbalanced datasets. The LINEX loss function, recognized for its asymmetric properties, effectively addresses misclassifications by treating them differently based on their characteristics, thereby improving the accuracy of ER in imbalanced scenarios. Additionally, the model is specifically designed to enhance the identification and matching of entities in datasets where certain classes are underrepresented. By utilizing a Primal-Dual optimization method, the model ensures robustness against noisy and imbalanced data while preserving the core structure of the LF-SSVM. Extensive experiments conducted on fifteen real-world ER datasets confirm the effectiveness of our approach, demonstrating substantial improvements over traditional methods in handling imbalanced class distributions in ER tasks.