Learning User Embeddings from Emails

摘要

Many important email-related tasks, such as email classification or search, highly rely on building quality document rep resentations (e.g., bag-of-words or key phrases) to assist matching and under standing. Despite prior success on rep resenting textual messages, creating qual ity user representations from emails was overlooked. In this paper, we propose to represent users using embeddings that are trained to reflect the email communi cation network. Our experiments on En ron dataset suggest that the resulting em beddings capture the semantic distance be tween users. To assess the quality of em beddings in a real-world application, we carry out auto-foldering task where the lexical representation of an email is en riched with user embedding features. Our results show that folder prediction accu racy is improved when embedding fea tures are present across multiple settings.