As I evaluate content on the network, I classify it and rate it. My identity associated with those things is established though (possibly) anonymous. Over time, islands of trusts within the web will form that can be used to help filter large amounts of content.
If I start a node, I can link to islands of trust to only allow verified content acceptable to my filters to pass through my resources.
It's not perfect. Some will attempt to game the system by building up trust and then attempting to sneak content through. Some will attempt to hide illicit content in innocuous content.