An Ensemble Approach to Cross-Domain Authorship Attribution

摘要

This paper presents an ensemble approach to cross-domain authorship attribution that combines predictions made by three independent classifiers, namely, standard character n-grams, character n-grams with non-diacritic distortion and word n-grams. Our proposal relies on variable-length n-gram models and multinomial logistic regression to select the prediction of highest probability among the three models as the output for the task. The present approach is compared against a number of baseline systems, and we report results based on both the PAN-CLEF 2018 test data, and on a new corpus of song lyrics in English and Portuguese.