Binary Cybergenre Classification Using Theoretic Feature Measures

An Examination of Genre Attributes for Web Page Classification

“…Previously reported results of this research project [4] addressed the issues of classifiers, feature sizes and feature selection measures for this classification problem. It found that in a binary classification scenario, the results using a Naïve Bayes classifier were as good as those obtained using a neural net classifier or a support vector machine.…”

“…On this basis [4] the research reported in this paper is based on using the Naïve Bayes classifier and the Information Gain Measure for feature selection. In addition, the results presented in this paper used the Multivariate Bernoulli Model form of the Naïve Bayes classifier as it significantly out performed the Multinomial Model.…”

“…A classification architecture was created consisting of 4 two-way classification subsystems, as per [4]. Each subsystem was trained to classify one of the 4 genres: Personal Home Page, FAQ, E-Shopping, and News.…”

“…Beyond 100 features, the results slowly decreased while interesting results were obtained at 5, 20 and 100 numbers of features. The features selected were those with the highest Information Gain values [4]. Experiments for 5, 20 and 100 features were run for each of the following 7 sets of attribute type combinations:…”

Open-Set Classification for Automated Genre Identification

“…frequency-based, chi-square, information gain, mutual information) and the classification model (e.g., SVM, decision trees, neural networks, etc.) [3,5,6,10,11,9,12,7]. To the best of our knowledge, all published studies consider AGI as a closed-set classification approach.…”