Meet Your Email Sender - Hybrid Approach to Email Signature Extraction

Abstract

Email signature is considered imperative for effective business email communication. Despite the growth of social media, it is still a powerful tool that can be used as a business card in the online world which presents all business information including name, contact number and address to recipients. Signatures can vary a lot in their structure and content, so it is a great challenge to automatically extract them. In this paper we present a hybrid approach to automatic signature extraction. First step is to obtain the original most recently sent message from the entire email thread, cleaned from all disclaimers and superfluous lines, making the signature to be at the bottom of the email. Then we apply Support Vector Machine (SVM) Machine Learning (ML) technique to classify emails according to whether they contain a signature. To improve obtained results we apply a set of sophisticated Information Extraction (IE) rules. Finally, we extract signatures with a great success. We trained and tested our technique on a wide range of different data: Forge dataset, Enron with our own collection of emails and a large set of emails provided by our native English-speaking friends. We extracted signatures with precision 99.62% and recall 93.20%.