Abstract

The World Wide Web has rich source of voluminous and heterogeneous information which continues to expand in size and complexity. Many Web pages are unstructured and semi-structured, so it consists of noisy information like advertisement, links, headers, footers etc. This noisy information makes extraction of Web content tedious. Many techniques that were proposed for Web content extraction are based on automatic extraction and hand crafted rule generation. Automatic extraction technique is done through Web page segmentation, but it increases the time complexity. Hand crafted rule generation uses string manipulation function for rule generation, but generating those rules is very difficult. A hybrid approach is proposed to extract main content from Web pages. A HTML Web page is converted to DOM tree and features are extracted and with the extracted features, rules are generated. Decision tree classification and Naïve Bayes classification are machine learning methods used for rules generation. By using the rules, noisy part in the Web page is discarded and informative content in the Web page is extracted. The performance of both decision tree classification and Naïve Bayes classification are measured with metrics like precision, recall, F-measure and accuracy.