Neo4j is ideal for this:
Adobe used 125 MongoDB servers to run their activity feed. It was replaced by 48 Cassandra servers. Now it runs on 3 (THREE!) servers of Neo4j. With more data (yet smaller disk footprint), higher load and more functionality. The scalability of a native graph database.
I personally built a social network for a startup, specifically for musicians and music lovers, it has 400,000 users and runs on one very stable server.
User accounts, their likes, their posts and so forth - this is an organic network/graph structure. We can benefit from the native graph databases path finding abilities and non index adjacency here.
As another example, I read a study recently, about how a startup was able to reduce the cost of their activity feed, for users of a temporal inbox, by using Amazon DynamoDB, because writes are cheap on this platform. I couldn't help but think that with Neo4j it would only be necessary to create a pointer to the content. I can't find that article right now, however, I'll try to dig it up.
