Graph-Based Realtime Recommendations for a Social Network

An early stage social network was marketing effectively and gaining users at a steady pace, but engagement was low and churn was rising. Product analytics indicated that users were unable to effectively discover each other, leading to a mistaken impression that the network contained little of relevance to them. Extrasensory¹ was engaged to correct this impression by putting personalized, relevant, and timely communications and recommendations in front of users across channels and product surfaces.

Motivation

Users of the social network were interested in connecting with new people that shared their interests and in being notified of relevant activities. As it stood, suggestions for new connections and notifications were not targeted, with activities being broadcast to the entire network and suggested connections being presented at random. This led to an unsustainable level of superfluous communication which was frustrating users and compounding by the day.

Approach

Extrasensory began by working through user personas and activities with product and marketing functions. These discussions made it clear that a system similar to the one Twitter (now X) used for recommending connections and content would be ideal. We designed and implemented a solution from the ground up as follows:

• Review of Twitter (now X) white papers and open source repositories to ascertain the state-of-the-art.
• Specification of a framework to define recommendation use cases according to user and business intentions.
• Development of an initial feature set, including "Who To Follow" recommendations and activity-based notifications.
• Construction of a purpose-built realtime social graph analytics engine to handle current and foreseen use cases.
• Automation of machine learning optimization using impression and engagement feedback from every recommendation.
• Integration with the app through a simple API in the manner of a 3rd party vendor.

Results

With the new recommender system built, deployed, and tested, engineering was able to integrate it easily and seamlessly. Users immediately noticed the improvement and responded with dramatically increased engagement. Product analytics indicated the users were successfully finding relevant activities and connections at greatly improved rates, and the total volume of user communication was put on a sustainable scaling trajectory. Excited by these successes, management began brainstorming additional uses to add using the extensible framework.