If you're looking for a project, I think this is something that an LLM, even a dumb local one, would be pretty good at. Give it a list of 50 articles you like, 50 you hate (or however many fit into the context window), and let it read the full text of each post and assign a 1-5 score. Then sort and/or filter by that.

In theory, this is actually a very textbook ML supervised learning problem, and stuffing an already-trained LLM's context window with a small handful of samples like I suggested is a gross hack. But it might be the easier option.