Results could be better filtered. SEO spam isn't very nuanced, it all reads the same and frequently plagiarized. I wonder if you can crudely use something like turnitin to filter for original content.

Maybe you could try and make a model SEO article for your own search engine, just taking your existing SEO results and figuring out which parameters are contributing the most to their ranking, then filtering out results that contain these parameters that worked well in your model. Rinse and repeat as SEO writers try and step up their arms race, but they should always end up being foiled by your changes to the search engine after optimizing your own perfect SEO model regularly.