Database and Expert Systems Applications

Many machine generated emails carry important information which must be acted upon at scheduled time by the recipient. Thus, it becomes a natural goal to automatically extract such actionable information from these emails and communicate to the users. These emails are generated for many different domains, providing different types of services. However, such emails carry personal information, therefore, it becomes difficult to get access to large corpus of labeled data for supervised information extraction methods. In this paper, we propose a novel method to automatically identify part of the email containing actionable information, called core region of the email, with the aid of a domain dictionary. Domain dictionary is generated based on the public information of the domain. The core regions are stored as template trees a template tree is a sub-tree embedded in the email’s HTML DOM tree. Our experiments over real data show, structure of the core region of the email, containing all the information of our interest, is very simple and it is 85%–98% smaller compared to the original email. Further, our experiments also show that the template trees are highly repetitive across diverse set of emails from a given service provider.