Activity Modelling using Email and Web Page Classification

This work explores the modelling of a user’s current activity using a single document and a very small collection of classified documents. A model called WeMAC was developed, which accretes evidence from heterogeneous sources to give a final classification of the user’s activity. The evidence sources considered here are the different attributes of a document. We evaluate the WeMAC model using two different document types: emails and web pages; assess its performance on both tiny document sets and larger sets; and assess its performance against a “one bag” approach. The WeMAC model compared well with existing results for systems with similar tasks and larger datasets, with an average F1 value of 0.5-0.7.