Seminar — Danushka Bollegala

In text mining scenarios, we often encounter the situation where we must apply a tool that was trained using a dataset that is different from the domain to which we are trying apply it. Considering the vast number of different domains in which we must apply a text mining tool and the costs involved in manually annotating data for each domain of interest to train a tool, it is attractive if we can somehow adapt a tool to each new domain of interest with minimum supervision. The fields of domain adaptation and transfer learning have gained a wide attention lately within this context. In this talk, I will present my work on domain adaptation for three tasks in NLP: a) cross-domain sentiment classification, b) transfer learning for semantic relation extraction, and c) cross-domain distributional semantics.

Danushka Bollegala is a Senior Lecturer at the Department of Computer Science at the University of Liverpool. Prior to joining the University of Liverpool, he was an Assistant Professor at the University of Tokyo, Japan. He obtained his PhD in Information Sciences from the University of Tokyo in 2009. His research interests are Natural Language Processing, Machine Learning, and Data Mining. He has worked in the past and currently on several problems in those fields such as multi-document summarisation, attributional and relational similarity measurement, name disambiguation, sentiment analysis, and relation prediction. He has published his work on conferences related to those fields such as ACL, EMNLP, WWW, IJCAI, and AAAI.