Scholarlink

<p>Web spam persistently undermines search engine integrity through deceptive practices such as cloaking and link manipulation, incurring substantial economic costs. Existing detection approaches often contend with significant challenges, including the effective management of uncertainty in human judgments and the processing of high-dimensional feature spaces. To address these limitations, we propose Graph-Augmented Web Spam Evidential Reasoner (GAWSER). This innovative framework integrates Dempster–Shafer Theory (DST) for robust uncertainty quantification with advanced graph-based feature selection. This system dynamically assesses judge reliability through adaptive evidence aggregation while employing spectral clustering on a feature similarity graph to achieve noticeable dimensionality reduction without compromising discriminative power. Evaluated on the WEBSPAM-UK2006 and WEBSPAM-UK2007 datasets, GAWSER demonstrated notable performance improvements, achieving accuracy increases of 4.2% (to 0.952) and 3.2% (to 0.947), respectively, alongside a reduction in false negatives. The framework consistently demonstrates improved metrics over both traditional and deep learning-based methodologies, offering a novel solution for uncertainty-aware web spam detection. Furthermore, recognizing the inherent scalability challenges posed by massive, dynamic web datasets, GAWSER’s design is specifically engineered to leverage high-performance computing (HPC) paradigms. Its graph-based feature selection and evidential reasoning components are inherently adaptable for parallel and distributed processing, which is critical for real-time, web-scale spam detection. This architectural consideration renders GAWSER highly relevant to the field of Supercomputing, showcasing a novel application for advanced computational systems in mitigating web spam.</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

Graph-augmented web spam detection using evidential reasoning

  • 期刊论文

  • 发表日期：

    2025年8月29日

查看全文
  • Amir Hosein Keyhanipour

摘要

Web spam persistently undermines search engine integrity through deceptive practices such as cloaking and link manipulation, incurring substantial economic costs. Existing detection approaches often contend with significant challenges, including the effective management of uncertainty in human judgments and the processing of high-dimensional feature spaces. To address these limitations, we propose Graph-Augmented Web Spam Evidential Reasoner (GAWSER). This innovative framework integrates Dempster–Shafer Theory (DST) for robust uncertainty quantification with advanced graph-based feature selection. This system dynamically assesses judge reliability through adaptive evidence aggregation while employing spectral clustering on a feature similarity graph to achieve noticeable dimensionality reduction without compromising discriminative power. Evaluated on the WEBSPAM-UK2006 and WEBSPAM-UK2007 datasets, GAWSER demonstrated notable performance improvements, achieving accuracy increases of 4.2% (to 0.952) and 3.2% (to 0.947), respectively, alongside a reduction in false negatives. The framework consistently demonstrates improved metrics over both traditional and deep learning-based methodologies, offering a novel solution for uncertainty-aware web spam detection. Furthermore, recognizing the inherent scalability challenges posed by massive, dynamic web datasets, GAWSER’s design is specifically engineered to leverage high-performance computing (HPC) paradigms. Its graph-based feature selection and evidential reasoning components are inherently adaptable for parallel and distributed processing, which is critical for real-time, web-scale spam detection. This architectural consideration renders GAWSER highly relevant to the field of Supercomputing, showcasing a novel application for advanced computational systems in mitigating web spam.