A link and Content Hybrid Approach for Arabic Web Spam Detection

摘要

Some Web sites developers act as spammers and try to mislead the search engines by using illegal Search Engine Optimizations (SEO) tips to increase the rank of their Web documents, to be more visible at the top 10 SERP. This is since gaining more visitors for marketing and commercial goals. This study is a continuation of a series of Arabic Web spam studies conducted by the authors, where this study is dedicated to build the first Arabic content/link Web spam detection system. This Novel system is capable to extract the set of content and link features of Web pages, in order to build the largest Arabic Web spam dataset. The constructed dataset contains three groups with the following three percentages of spam contents: 2%, 30%, and 40%. These three groups with varying percentages of spam contents were collected through the embedded crawler in the proposed system. The automated classification of spam Web pages used based on the features in the benchmark dataset. The proposed system used the rules of Decision Tree; which is considered as the best classifier to detect Arabic content/link Web spam. The proposed system helps to clean the SERP from all URLs referring to Arabic spam Web pages. It produces accuracy of 90.1099% for Arabic content-based, 93.1034% for Arabic link-based, and 89.011% in detecting both Arabic content and link Web spam, based on the collected dataset and conducted analysis.