Novel Approach for Tagging of Discourse Segments in Help-Desk E-Mails

The volume of email that help-desks receive every day is very high and often queries are repeated. Any kind of automation in processing of emails requires good understanding of the emails. In the current work we propose a schema for tagging author composed sentences in help-desk emails by the intent of the author. We have created a corpus taking email data from two help-desks and annotated them at sentence level. We have achieved significant accuracies in learning to automatically tag the sentences using n-gram features and some hand-picked lexical features. At every stage right from choice of schema to choice of features, we have tried to be domain independent or keep domain related information as a separate component. Automation of Tagging of Discourse Segments (TODS) in email, we propose is a significant step towards finding the discourse parse of emails.