Content-based analysis to detect Arabic web spam

Abstract

Search engines are important outlets for information query and retrieval. They have to deal with the continual increase of information available on the web, and provide users with convenient access to such huge amounts of information. Furthermore, with this huge amount of information, a more complex challenge that continuously gets more and more difficult to illuminate is the spam in web pages. For several reasons, web spammers try to intrude in the search results and inject artificially biased results in favour of their websites or pages. Spam pages are added to the internet on a daily basis, thus making it difficult for search engines to keep up with the fast-growing and dynamic nature of the web, especially since spammers tend to add more keywords to their websites to deceive the search engines and increase the rank of their pages. In this research, we have investigated four different classification algorithms (naïve Bayes, decision tree, SVM and K-NN) to detect Arabic web spam pages, based on content. The three groups of datasets used, with 1%, 15% and 50% spam contents, were collected using a crawler that was customized for this study. Spam pages were classified manually. Different tests and comparisons have revealed that the Decision Tree was the best classifier for this purpose.

A link and Content Hybrid Approach for Arabic Web Spam Detection

“…The study of [16] improves their previous studies on the content-based Arabic Web spam. They used a large Arabic content-based spam dataset which contains 15,000 Web pages, that were collected by a special crawler.…”

“…All the previous Arabic Web spam studies [12][13][14][15][16][17] tried to identify the best classification algorithm for the content-based Arabic Web spam detection, which almost unanimously indicate the Decision Tree classifier is the best. Therefore [18] based on the 15,000 Arabic spam Web pages, enhanced more content-based features, and built the novel Arabic Web spam detection system using the rules of Decision Tree classification algorithm.…”

“…In this paper we have built a large Arabic Web spam dataset containing around 23,000 Web pages, where 18,000 Web pages of the total Web spam dataset were used as training dataset which extended the datasets mentioned in [12][13][14][15][16][17][18][19]. The rest of the Web spam dataset consists of 5,000 Web pages that were used as test dataset.…”

“…We developed a Web page analyzer capable to extract the previous proposed features by [12][13][14][15][16][17][18][19], and proposed new content ad link features.…”

“…In addition, the researches so far are few. So this paper adopts an enhancement to the previous studies of contentbased/link-based Arabic Web spam detection [12][13][14][15][16][17][18][19]. We collected a larger Arabic Web spam dataset to improve the Arabic Web spam features mentioned in the previous studies, and built a content/link Arabic Web spam detection system.…”