I am optimistic that a social graph can help with this, using transitive trust and making it easy to transitively "cut off" bad parts of the graph with a blocklist. The problem is that commercial graphs do become useless, as you say.
But with the Fediverse we now have a graph that is resistant to this. I'm not aware of anyone building a reputation system on top of it, but this is something I expect will develop in the future. The system's opinion about a particular data point would then be personalized based on the people you say you trust to indicate that information, combined with the people they say they trust, and so forth.
But with the Fediverse we now have a graph that is resistant to this. I'm not aware of anyone building a reputation system on top of it, but this is something I expect will develop in the future. The system's opinion about a particular data point would then be personalized based on the people you say you trust to indicate that information, combined with the people they say they trust, and so forth.