EPSMS and the Document Occurrence Representation for Authorship Identification - Notebook for PAN at CLEF 2011

This paper describes the participation of the PISIS team in the authorship identification track of PAN’11. We adopted two different strategies for the tasks of authorship attribution and authorship verification. For authorship attribution we performed experiments with a document occurrence representation using a standard classification-based approach. Results obtained with this approach were mixed: in the small data sets distributional representations resulted very helpful, although in the large data sets a simple bag-of-words approach outperformed the document occurrence approach. For authorship verification we adopted a classification-based approach and proposed a modification to Ensemble Particle Swarm Model Selection (EPSMS) for selecting classification models for each task. This approach obtained acceptable performance in two out of the three data sets.