Author Profiling for English and Arabic Emails08 February 2010
This paper reports on some aspects of a research project aimed at automating the analysis of texts for the purpose of author profiling and identification. The Text Attribution Tool (TAT) was developed for the purpose of language-independent author profiling and has now been trained on two email corpora, English and Arabic. The complete analysis provides probabilities for the author’s basic demographic traits (gender, age, geographic origin, level of education and native language) as well as for five psychometric traits. The prototype system also provides a probability of a match with other texts, whether from known or unknown authors. A very important part of the project was the data collection and we give an overview of the collection process as well as a detailed description of the corpus of email data which was collected. We describe the overall TAT system and its components before outlining the ways in which the email data is processed and analysed. Because Arabic presents particular challenges for NLP, this paper also describes more specifically the text processing components developed to handle Arabic emails. Finally, we describe the Machine Learning setup used to produce classifiers for the different author traits and we present the experimental results, which are promising for most traits examined.