This repository contains all scripts associated with my research on topical Web-page classification. You can read the full paper describing the task, experiments, and results here.
Multi-class topical web-page classification is a difficult task with widespread application. Throughout this paper, I analyze the performance of well-studied techniques on two different representations of web-pages: hand-written meta-descriptions and on-page text content. I acquired all of the training labels and website descriptions from the DMOZ dataset and all of the on-page content from scraping the actual web-pages. I achieved 74.035% and 79.121% accuracy for on-page content and website descriptions respectively in a 16-way classification task with a 42.032% most frequently tagged baseline accuracy.