Abstract:

As the number of Web sites is growing rapidly, classifying Web pages with respect to their contents proposes itself as a possible solution to prevent accessing malicious content that may be found on these sites or to access useful information in a easier way. With such a classification, access to specific sites may be allowed or these sites may be filtered and thus access to them may be prevented. In this study, the Web site classification problem is examined by using different machine learning methods and artificial neural networks. In order to solve this classification problem, two different approaches are proposed, namely Binary Classification and Multiple Classification. Both approaches are tested and their performance are compared by using a number of Web sites collected for this study. Considering all experimental results, it has been found that the Binary Classification approach is more effective only when it is used to perform the task of filtering a desired website class. In terms of performance, Logistic Regression is the best performance algorithm for binary classifiers. Among the algorithms applied in the Multiple Classification approach, Support Vector Machines (SVM) is found as the most successful method. Furthermore, different word vectorization methods have been employed and their performance have been compared within the Multiple Classification problem. Algorithms used in Binary and Multi-class Classification approaches have been separately tested using different vectorization methods. By this way the classification and content filtering problems on Web pages have been approached together, thus differentiating this study from similar researches in the domain.

Keywords: