Shih-Wen Ke’s research while affiliated with University of Sunderland and other places

Improving the accuracy of assigning new email messages to small folders can reduce the likelihood of users creating duplicate folders for some topics. In this paper we presented a hybrid classification model, PERC, and use the Enron Email Corpus to investigate the performance of kNN, SVM and PERC in a simulation of a real­time situation. Our results show that PERC is significantly better at assigning messages to small folders. The effects of dif­ ferent parameter settings for the classifiers are discussed.

This paper describes the text mining of personal document collections in order to learn the categories of the documents in the collection, and to assign a suitable text label to each category. In the first experiment we make use of a pre­ classified collection of documents from which we extract a text label for each category. In the second experiment we use the k­means clustering algorithm to automatically learn which categories are present in a collection, then generate short text labels to describe each category. The technique, which is essentially a form of automatic indexing, can be used whenever class exemplars are generated as part of the document clustering process.