Authorship Attribution Author Style vs Content

The problem of authorship attribution has long been debated and remains a hot topic. This discipline has seen remarkable progress in the previous decade as a result of breakthroughs in digital computers. Presented study is one of it’s kind and brings a fresh perspective to investigate the relevance and significance of style-based vs content-based authorship attribution using two corpora of presidential campaign text media with varying levels of topic heterogeneity. Using the same model trained with content-based features as a baseline, it was discovered that a model with style-based features performs better on topic-homogeneous documents than topic-heterogeneous documents. It was also interesting to see that models with style-based featurization generalize more effectively than models with content-based featurization across corpora with varying degrees of topic heterogeneity. These findings show how understanding the content of the underlying texts can help when picking attributes for authorship attribution. Also highlighting some key discoveries along the way.