Mining mailing lists for content

Abstract

In large decentralized institutions such as MIT, finding information about events and activities on a campus-wide basis can be a strenuous task. This is mainly due to the ephemeral nature of events and the inability to impose a centralized information system to all event organizers and target audiences. For the purpose of advertising events, Email is the communication medium of choice. In particular, there is a wide-spread use of electronic mailing lists to publicize events and activities. These can be used as a valuable source for information mining. This dissertation will propose two mining architectures to find category-specific event announcements broadcasted on public MIT mailing lists. At the center of these mining systems is a text classifier that groups Emails based on their textual content. Classification is followed by information extraction where labeled data, such as the event date, is identified and stored along with the Email content in a searchable database. The first architecture is based on a probabilistic classification method, namely naive-Bayes while the second uses a rules-based classifier. A case implementation, FreeFood@MIT, was implemented to expose the results of these classification schemes and is used as a benchmark for recommendations.