CNG Text Classification for Authorship Profiling Task Notebook for PAN at CLEF 2013.
Magdalena Jankowska, Vlado Kešelj, Evangelos Milios · CLEF (Working Notes) · 2013
We describe our participation in the Author Profiling task of the PAN 2013 competition. The task objective is to determine the age and the gender of an author of a document. We applied the Common N-Gram (CNG) classifier (Keselj et al., 2003) to this task. The CNG classifier uses a dissimilarity measure based on the differences in the frequencies of the character n-grams that are most common in the considered documents. To train the classifier, a class is represented by one class document created by concatenating the training documents belonging to the class. A sample document is labelled by the class with the minimum dissimilarity. For the six class classification (combinations of two possible gender labels and three possible age labels) we achieved the accuracy of 0.2814 on the English test dataset and 0.2592 on the Spanish test dataset. Our results are below the medians of the results of the competition participants.