PAN at CLEF 2014

Recently I was asked to assist in a text attribution problem: the illegitimate reuse of text (and images) from the web pages of a small business. When the offender was approached about committing possible plagiarism their response was ?prove it?. This talk will describe how I approached the problem of proving ownership and the challenges it entailed. I will describe the experiences gained from working on the EPSRC-funded Measuring Text Reuse (METER) project with the UK Press Association, along with the current text attribution literature, that informed my textual analysis. I will also demonstrate how freely available resources can be used to tackle this kind of problem. In the end was I successful: did the offender admit their guilt? You?ll have to attend the talk to find out!

Personality recognition from text consists in the automatic classification of authors' personality traits from pieces of text they wrote. Classifier's predictions can be compared against gold standard labels, obtained by means of personality assessments like the Big5 personality test. Until recently, the extraction of personality types was limited to blogs and offline texts, while in recent years there is a strong interest in the scientific community about the extraction of personality from various sources, such as online social networks, speech and video. Current approaches to Personality Recognition are based on supervised learning, but this has several limitations, for example the cost of data annotation, the lack of domain adaptability and multilinguality. We present an unsupervised method for personality recognition from text and some of its applications in Social network analysis as well as in other NLP tasks.