Modeling heterogeneous embeddings from workplace communications

In the modern workplaces, people interact with each other in many different ways: email, chat, phone call, meeting, etc. By analyzing the user interactions in the Office products, we can offer useful insights our users. However, learning features from a large scale, dynamic, heterogeneous network is still a great challenge to the traditional network embedding methods. This project aims at learning the embeddings from both the content and the structure of the interactions to better utilize the heterogeneous data. The embeddings can be used for various recommendation tasks. For example, they can suggest email recipients based on the related meetings, or suggest to schedule a meeting based on the recent chats.