ABSTRACT
Mining useful information to analyze knowledge-intensive business processes requires data that describes activities of knowledge workers. Emails are widely used in organizations to provide support in the functioning of knowledge-intensive processes. The recent COVID-19 pandemic has increased reliance on technologies such as email to help facilitate communication within organizations to make up for the lack of face-to-face contact. In this work, we propose an activity mining technique, which receives an incoming email message, classifies the sender's intent and translates it into a set of business process activities. Specifically, we leverage deep learning language models to first classify the email body into a group of intents, which are then mapped to related activities. To our knowledge, we propose the first transfer-learning based solution for mining activity information from emails. The effectiveness of our solution was evaluated on real-world data coming from email exchanges between knowledge workers. Our results based on unsupervised experiments and a field study show that transformer models can be used to semantically label emails and that mapping activities to matched intents is highly accurate. © 2023, Springer Nature Switzerland AG.