Hierarchical Clustering Analysis: The Best-Performing Approach at PAN 2017 Author Clustering Task

The author clustering problem consists in grouping documents written by the same author so that each group corresponds to a different author. We described our approach to the author clustering task at PAN 2017, which resulted in the best-performing system at the aforementioned task. Our method performs a hierarchical clustering analysis using document features such as typed and untyped character n-grams, word n-grams, and stylometric features. We experimented with two feature representation methods, log-entropy model, and TF-IDF, while tuning minimum frequency threshold values to reduce the feature dimensionality. We identified the optimal number of different clusters (authors) dynamically for each collection using the Calinski Harabasz score. The implementation of our system is available open source (https://github.com/helenpy/clusterPAN2017).