Segmenting Email Message Text into Zones

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 919–928, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Segmenting Email Message Text into Zones Andrew Lampert †‡ †CSIRO ICT Centre PO Box 76 Epping 1710, Australia andrew.lampert@csiro.au Robert Dale ‡ rdale@ics.mq.edu.au C´ecile Paris † ‡Centre for Language Technology Macquarie University North Ryde 2109, Australia cecile.paris@csiro.au Abstract In the early days of email, widely-used conventions for indicating quoted reply content and email signatures made it easy to segment email messages into their func- tional parts. Today, the explosion of dif- ferent email formats and styles, coupled with the ad hoc ways in which people vary the structure and layout of their messages, means that simple techniques for identify- ing quoted replies that used to yield 95% accuracy now find less than 10% of such content. In this paper, we describe Zebra, an SVM-based system for segmenting the body text of email messages into nine zone types based on graphic, orthographic and lexical cues. Zebra performs this task with an accuracy of 87.01%; when the num- ber of zones is abstracted to two or three zone classes, this increases to 93.60% and 91.53% respectively. 1 Introduction Email message bodies consist of different func- tional parts such as email signatures, quoted re- ply content and advertising content. We refer to these as email zones. Many language process- ing tools stand to benefit from better knowledge of this message structure, facilitating focus on rel- evant content in specific parts of a message. In particular, access to zone information would al- low email classification, summarisation and anal- ysis tools to separate or filter out ‘noise’ and focus on the content in specific zones of a message that are relevant to the application at hand. Email con- tact mining tools such as that developed by Culotta et al. (2004), for example, might access the email signature zone, while tools that attempt to iden- tify tasks or action items in email (e.g., (Bellotti et al., 2003; Corston-Oliver et al., 2004; Bennett and Carbonell, 2007; Lampert et al., 2007)) might restrict themselves to the sender-authored and for- warded content. Despite previous work on this problem, there are no available tools that can re- liably extract or identify the different functional zones of an email message. While there is no agreed standard set of email zones, there are clearly different functional parts within the body text of email messages. For ex- ample, the content of an email disclaimer is func- tionally different from the sender-authored content and from the quoted reply content automatically included from previous messages in the thread of conversation. Of course, there are different dis- tinctions that can be drawn between zones; in this paper we explore several different categorisa- tions based on our proposed set of nine underlying email zones. Although we focus on content in the body of email messages, we recognise the presence of use- ful information in the semi-structured headers, and indeed make use of header information such …

Content truncated to 500 words