Author Identification Using Latent Dirichlet Allocation

We tackle the task of author identification at PAN 2015 through a Latent Dirichlet Allocation (LDA) model. By using this method, we take into account the vocabulary and context of words at the same time, and after a statistical process find to what extent the relations between words are given in each document; processing a set of documents by LDA returns a set of distributions of topics. Each distribution can be seen as a vector of features and a fingerprint of each document within the collection. We used then a Naive Bayes classifier on the obtained patterns with different performances. We obtained state-of-the-art performance for English, overtaking the best FS score reported in PAN 2015, while obtaining mixed results for other languages.