Application of Bridging to Semi-Supervised Text LearningDr Josiah Poon Tuesday 20th June 2006 at 11am
AbstractSemi-supervised learning is an attempt to improve over supervised learning by building a classifier from labelled data as well as the unlabelled data. For example, in the context of spam filtering, the labelled data consists of emails labelled as either spam or useful emails, while unlabelled data do not have these labels, but they are essentially emails. A method called Bridging has been shown to be helpful in a purely supervised setting where unlabelled data are being classified via an intermediate set of related, but different, data. The rationale behind this method is that there usually exists a great deal of related information in another format. For example, web pages of companies and those pages that interest a particular user may be helpful in spam filtering. The words typical in spam may be found in commercial sites, while information found in useful emails is likely to occur in what the user considers interesting websites. In this work, we convert bridging into a semi-supervised method in an attempt to generate a powerful semi-supervised algorithm. Short resumeDr. Josiah Poon is a Lecturer from the School of Information Technologies in the U. of Sydney. Before taking up this position, he was a faculty member in the Department of CSEE in the U. of Queensland, a research engineer with CISRA (Canon Information System Research Australia) and a system analyst with a major international bank. He holds a B.Sc(Hons) from U. of Manchester (UK), M.Sc. from Deakin University and a Ph.D from the University of Sydney. His Ph.D. research has contributed to the modelling of design process as co-evolution using modified genetic algorithms. His current interests include machine learning, data mining, pervasive computing and CS Education. He is also the coordinator of the Machine Learning Seminar in the School. |