These days it’s not about normal growth, it’s about finding and driving hockey-stick levels of growth. Sales and marketing organizations are looking to AI to help growth hack their way to new markets and new segments. We have used
Mutual Information for many years to help filter out noise and find the critical insights to new cohorts of users, devices, businesses, and networks and now we can do it at nearly infinite granularity across massive data sources and dimensions.
Random variables share information with each other. The amount of information shared between then is symmetric and computed via Mutual Information. If you know the temperature is 90 degrees you have information about what month of the year it might be. If the month is December there is information about what temperature to expect. We employed MI ranking in the past to distinguish devices used by teacher vs student, developer vs non-developer, and IT Professional vs non-IT professionals. This helps us make Windows better for the applications these key segments care about. The alternative is to build out labeled data for a custom ML model for each new segment.
In this presentation we will explain the problem space of fine grained segmentation, provide cross industry use cases, and demonstrate how the code works with a real world example. We will walk attendees through a detailed internal Microsoft case study for applying MI and share the actual code to include a copy capable of running on Azure Data Lake. We will show how the code can scale to billions of nodes and edges and even demonstrate some of the output in our Neo4J graph.