Scholarlink

<p>Online Social Networks (OSNs) have become a major part of everyday communication, but they are still exposed to several security problems. Most existing studies mainly focus on spotting external attacks, and there is far less work on identifying users who turn harmful from internal networks. In this study, a real time framework is introduced to monitor users by updating their trust levels continuously as their behavior changes. The model was tested on a dataset of 10,000 users. Three clustering methods, K-means, DBSCAN and Hierarchical Clustering, were used to group users as trusted, under observation, or intruders. Among these, K-means provided the most stable results with a silhouette score of 0.356, while DBSCAN flagged 353 users as outliers. Reliability of clusters was validated by using Random Forest and XGBoost with macro F-1 scores of 0.98 and 0.99, respectively. That implies clustering was done thoroughly. Additionally, changes in trust scores reflect changes in user behavior. The framework is evaluated using a semi-synthetic OSN dataset designed to emulate realistic behavioral and structural characteristics of online social platforms. The proposed framework introduces a novel approach that can be further extended and integrated into existing OSNs to enhance user security.</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

Dynamic trust based intrusion detection for internal threat mitigation in online social networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月11日

查看全文
  • Aparna Agarwal,
  • Gordhan Jethava,
  • Sweta Jethava

摘要

Online Social Networks (OSNs) have become a major part of everyday communication, but they are still exposed to several security problems. Most existing studies mainly focus on spotting external attacks, and there is far less work on identifying users who turn harmful from internal networks. In this study, a real time framework is introduced to monitor users by updating their trust levels continuously as their behavior changes. The model was tested on a dataset of 10,000 users. Three clustering methods, K-means, DBSCAN and Hierarchical Clustering, were used to group users as trusted, under observation, or intruders. Among these, K-means provided the most stable results with a silhouette score of 0.356, while DBSCAN flagged 353 users as outliers. Reliability of clusters was validated by using Random Forest and XGBoost with macro F-1 scores of 0.98 and 0.99, respectively. That implies clustering was done thoroughly. Additionally, changes in trust scores reflect changes in user behavior. The framework is evaluated using a semi-synthetic OSN dataset designed to emulate realistic behavioral and structural characteristics of online social platforms. The proposed framework introduces a novel approach that can be further extended and integrated into existing OSNs to enhance user security.