Jainesh Patel - Academia.edu (original) (raw)
Related Authors
University of Ottawa | Université d'Ottawa
Uploads
Papers by Jainesh Patel
As with the use of internet, use of emails increases drastically for electronic communication. Th... more As with the use of internet, use of emails increases drastically for electronic communication. This leads the mail boxes gets congested and emerged the problem of email overload, which is solved with the help of email categorization or email management. Email Categorization is multifaceted problem with many difficulties. Many schemes have been proposed for solving this problem in either supervised or unsupervised approach. With that approach once categorization model is built, it is hard to make any changes to them for handling of dynamic situations. As email replicates current information around the globe, the email content will be changed with the passage of time. Concept drift is the situation which occurs due to changes in underlying data distribution over a time period. The problem of concept drift detection and handling will occur due to dynamic nature of email. This paper proposes the dynamic hybrid scheme, combines supervised and unsupervised approach for detection and handling of concept drift. Initial classifier is built with the help of classification algorithm, and then clustering algorithm is applied in 'General' category of classifier to detect concept drift.. If it is detected then new cluster is formed for that new emerging concept and appropriate label is assigned to that cluster.
As with the use of internet, use of emails increases drastically for electronic communication. Th... more As with the use of internet, use of emails increases drastically for electronic communication. This leads the mail boxes gets congested and emerged the problem of email overload, which is solved with the help of email categorization or email management. Email Categorization is multifaceted problem with many difficulties. Many schemes have been proposed for solving this problem in either supervised or unsupervised approach. With that approach once categorization model is built, it is hard to make any changes to them for handling of dynamic situations. As email replicates current information around the globe, the email content will be changed with the passage of time. Concept drift is the situation which occurs due to changes in underlying data distribution over a time period. The problem of concept drift detection and handling will occur due to dynamic nature of email. This paper proposes the dynamic hybrid scheme, combines supervised and unsupervised approach for detection and handling of concept drift. Initial classifier is built with the help of classification algorithm, and then clustering algorithm is applied in 'General' category of classifier to detect concept drift.. If it is detected then new cluster is formed for that new emerging concept and appropriate label is assigned to that cluster.