This paper presents our approach to the Author Profiling (AP) task at PAN 2016. The task aims at identifying the author's age and gender under crossgenre AP conditions in three languages: English, Spanish, and Dutch. Our preprocessing stage includes reducing non-Textual features to their corresponding semantic classes. We exploit typed character n-grams, lexical features, and nontextual features (domain names). We experimented with various feature representations (binary, raw frequency, normalized frequency, second order attributes (SOA), tf-idf) and machine learning algorithms (liblinear and libSVM implementations of Support Vector Machines (SVM), multinomial naive Bayes, logistic regression). For textual feature selection, we applied the transition point technique, except when SOA was used. We found that the optimal configuration was different for different languages at each stage.
|Name||CEUR Workshop Proceedings|
|Conference||2016 Working Notes of Conference and Labs of the Evaluation Forum, CLEF 2016|
|Period||5/09/16 → 8/09/16|