Hybrid influence model improves spam webpage detection in web-network tests

Researchers report that a gravity-based hybrid model improved the discrimination of spam webpages in directed web graphs and two real-world web-spam datasets, WEBSPAM-UK2006 and WEBSPAM-UK2007. The proposed Hybrid Influence Gravity Model, or HIGM, combines weighted degree, improved weighted k-shell decomposition, Laplacian centrality, neighborhood structural influence, and average shortest-path variation. The study compared this combined measure with traditional centrality measures through several machine-learning approaches, including Naive Bayes, support vector machines and AdaBoost, as well as classification and regression models.

The authors report better spam discrimination, robustness and scalability, although the supplied article summary does not provide individual accuracy scores, dataset sizes, error rates or the size of the performance advantage. The research was presented in Frontiers in Applied Mathematics and Statistics as a computational study of spam-page detection in directed web networks. Its central concern is the limited information captured by conventional centrality measures when a webpage’s spam-related behavior depends on interactions within a neighborhood and across the wider link structure.

Measures including degree, closeness, betweenness, eigenvector, eccentricity, Laplacian, DKE and DKEGM centralities are described by the authors as focusing primarily on isolated structural characteristics. HIGM was designed to combine local and global network information in a single structural measure, with the aim of capturing both a webpage’s position in the graph and patterns relevant to spam propagation. Technically, HIGM integrates five components named in the study: weighted degree, improved weighted k-shell decomposition, Laplacian centrality, neighborhood structural influence and average shortest-path variation.

Together, these signals allow the proposed model to assess a webpage through a broader set of structural properties than any one of the centrality measures used in the comparison. The paper presents this integration as a way to reflect higher-order neighborhood interactions and the behavior associated with spam spreading in complex web graphs. On top of the combined structural score, the researchers introduce a gravity-inspired spam-influence function.

This component models interaction over multiple links and is intended to represent spam-propagation behavior within the network. HIGM also incorporates clustering-based penalization, which the authors designed to lessen the effect of artificially dense spam structures. In comparative experiments, the framework was initially demonstrated on a directed web graph and then evaluated with the WEBSPAM-UK2006 and WEBSPAM-UK2007 datasets.

Relative to the traditional centrality baselines, the study reports improved ability to distinguish spam webpages, together with gains described as robust and scalable. These are results from the authors’ computational evaluations, not evidence that the method has been deployed by an operational web platform or tested outside the reported graph and datasets. The findings remain bounded by the scope of the study.

The article evidence describes a structural classification approach applied to directed web graphs, but it does not establish performance in other kinds of networks, under changing spam behavior or beyond the specific test settings. It also does not quantify the contribution of each HIGM component, disclose the individual classifier results in the supplied summary, or show that gravity-inspired influence is uniquely effective for a particular category of spam. What the study does report is a unified model that joins several graph measures, multi-hop influence and clustering-based adjustment, then uses the resulting score with established machine-learning methods to improve spam-page discrimination in its comparative tests.

The work was authored by DSDivya S. Raj and Jaganathan B of the Vellore Institute of Technology–Chennai Campus in India; the publisher’s record stated that the final formatted article was forthcoming when the research record was issued.

Subject of Research: Graph-based spam webpage detection in directed web networks.

Article Title: An improved gravity-based hybrid influence model for spam webpages detection using Laplacian centrality, improved weighted k-shell, and clustering frameworks

Article References: Raj, D. S., & B., J. (2026). An improved gravity-based hybrid influence model for spam webpages detection using Laplacian centrality, improved weighted k-shell, and clustering frameworks. Frontiers in Applied Mathematics and Statistics, 12, Article 1930812. https://doi.org/10.3389/fams.2026.1930812

Image Credits: AI Generated

DOI: 10.3389/fams.2026.1930812

Keywords: spam webpage detection, web graph analysis, gravity-based model, Laplacian centrality, weighted k-shell, network clustering, web spam, machine learning, directed networks, network influence, improved, gravity-based