contact: izavits@iit.demokritos.gr
In various applications, there is a need to identify objects of a specific class amongst all objects by learning from a training set containing labels of the objects of that class. This setting is known as one-class classification, unary classification, or class modeling. A similar setting is PU learning (Positive – Unlabeled learning), in which a binary classifier is learned semi-supervised from only positive and unlabeled samples. This project will touch on the intersection of those settings, trying to build comparison methods and efficient document representations to facilitate document classification in various domains. After the successful undertaking of the project, the students will be able to describe what one-class classification and PU learning is about, use NLP tools to analyze texts, use machine learning to represent texts, and use data mining techniques to identify texts belonging to the desired class.