Content based web page categorization

Classification of Web Pages is a machine learning problem which gets more and more important every day. Categorizing web pages provides useful information for efficient internet use, spam filtering and many other application areas. Finding results quickly from the millions of websites users are looking for is a problem that must be solved for search engines. Web page classification can also be used by cyber security applications by blocking web pages with malicious content before they are displayed by the user. Web page classification is an Information Retrieval application that provides useful information that can be a basis for many different application domains. Another example of application is the creation of an internet usage profile of a user for network anomaly detection. In this study, a system for classifying web pages was developed. Deep learning-based approaches have been tested and used in the developed system. Textual information in the content of the web page is used to classify web pages. For the classification mechanism, the meta tags contained in the web page are used. The meta tags used for classification are title, description, keywords. During the tests, a deep learning architecture based on Recursive Neural Networks (RNN) was used. Performance analysis has been performed by performing some hyperparameter tuning on this deep learning architecture. In addition, Transfer Learning has been tested in the developed system. Transfer Learning is the name given to the approach of building a machine learning model using pre-trained parameters to solve a problem. According to the results obtained, the success rate of the web page classification system is about 85%. The tests were run on the CPU and GPU, and the performance comparison of the run times obtained on this two different hardware was made.