Authorship attribution of Spanish poems using n-grams and the web as corpus

Abstract

In many areas of professional development, the categorization of textual objects is of critical importance. A prominent example is the attribution of authorship, where symbolic information is manipulated using natural language processing techniques. In this context, one of the main limitations is the necessity of a large number of pre-labeled instances for each author that is to be identified. This paper proposes a method based on the use of n-grams of characters and the use of the web to enrich the training sets. The proposed method considers the automatic extraction of the unlabeled examples from the Web and its iterative integration into the training data set. The evaluation of the proposed approach was done by using a corpus formed by poems corresponding to 5 contemporary Mexican poets. The results presented allow evaluating the impact of the incorporation of new information into the training set, as well as the role played by the selection of classification attributes using information gain.