Abstract
With online calendar services gaining popularity worldwide, calendar data hasbecome one of the richest context sources for understanding human behavior.However, event scheduling is still time-consuming even with the development ofonline calendars. Although machine learning based event scheduling models haveautomated scheduling processes to some extent, they often fail to understandsubtle user preferences and complex calendar contexts with event titles writtenin natural language. In this paper, we propose Neural Event SchedulingAssistant (NESA) which learns user preferences and understands calendarcontexts, directly from raw online calendars for fully automated and highlyeffective event scheduling. We leverage over 593K calendar events for NESA tolearn scheduling personal events, and we further utilize NESA formulti-attendee event scheduling. NESA successfully incorporates deep neuralnetworks such as Bidirectional Long Short-Term Memory, Convolutional NeuralNetwork, and Highway Network for learning the preferences of each user andunderstanding calendar context based on natural languages. The experimentalresults show that NESA significantly outperforms previous baseline models interms of various evaluation metrics on both personal and multi-attendee eventscheduling tasks. Our qualitative analysis demonstrates the effectiveness ofeach layer in NESA and learned user preferences.