Boundary detection of multiple related temporal duration of schedules in email

Abstract

Emails are very popular method for information exchange between people. In this paper, an approach to annotate the starting time (stime) and ending time (etime) of duration in schedule notices is proposed. Most related works have reported on only seminar announcements, most of which contain only one schedule per announcement and are written in very restricted format. Different from those seminar announcements, an email frequently contains information about multiple schedules with highly complex format. To process the emails, the proposed system first detects and normalizes all time expressions of the email using regular expression patterns, and then determines which time expression actually represents stime and etime information of schedules. Evaluation is carried out on newly constructed Korean email corpus, and it shows 87.35 % of F1-score for stime and 85.13 % for etime.