No abstract

Web spam is one of the most important problems which degrade quality and efficiency of web search engines. In this paper, we present a novel link-based ant colony optimization learning algorithm for spam host detection. The host graph is first constructed by aggregating pages' hyperlink structure. Following the TrustRank assumption, ants start walking from a normal host and randomly follow host links with a probability distribution. Then, the classification rules are appropriately generated according to common features of normal hosts sequentially discovered by ants. From the experiments with the WEBSPAM-UK2006 dataset, the proposed learning model provides much accuracy in classifying both normal and spam hosts than several baselines, including a state of the art C4.5. Moreover, we also provide an analysis in parameter tuning for better results.