Context-Aware Intent Identification in Email Conversations

摘要: Email continues to be one of the most important means online communication. People spend a significant amount time sending, reading, searching and responding email in order manage tasks, exchange information, etc. In this paper, we study intent identification workplace email. We use large scale publicly available dataset characterize intents enterprise propose methods for improving conversations. Previous work focused on classifying messages into broad topical categories or detecting sentences that contain action items follow certain speech acts. work, focus sentence-level how incorporating more context (such as full message body other metadata) could improve performance models. experiment with several models leveraging including both classical machine learning deep approaches. show modeling interaction between sentence can significantly performance.