System for email analyzing and response generation based on machine learning

Abstract

Email is one of the most popular and effective communication channels in the modern world both for personal tasks and for business. However, the amount of unanswered received emails sometimes can be overwhelming. Systems for automatic email response generation and analysis through unsupervised machine learning technics have recently been proposed and are getting more popular. Unfortunately, these systems mostly partially provide options that work great in services like chat-bots, but not in the business correspondence tasks. The main purpose of the dissertation is to provide an open-source system based on deep learning neural networks, which would be capable of generating automatic replies based on the user’s personal history of email replies. The system is based on a sequence-to-sequence encoder-decoder approach with long short-term memory architecture.