We may introduce some bias through our algorithm, insofar as the weights we assign various factors favor one type of article over another. But we haven't used a training set yet (which is where ingrained biases would be factored in) so the algorithm is fairly simple for now.
I agree with you this isn't an easy problem to crack. We wanted to see if (a) there's interest in a solution and (b) reservations about our approach before investing more. So this is exactly the kind of feedback that's helpful. Thank you again.
I'm not heavy into the soft sciences. I'm a mathematician (complete with Ph.D.!) if that'll help you? I'm not sure if you're open source. To be truly objective, I may actually need to see that.
I'm very much retired - and willing to sign both an NDA and a non-compete, if you want. If you don't want, I'm not sure how much I can offer other than observations. It will also take a bunch of research to the methodology, but I'm not actually doing anything better.
It does make me question this, however. How are you certain you're not introducing your own biases with it? Or, perhaps not even biases of your own?
Sometimes bias exists simply by virtue of externalities. An example might be that, when differentiated by economic ability, black people don't actually do more drugs than white people - though crime stats would indicate otherwise. The difference is their neighborhoods are policed more heavily and the biases in enforcement. Notably, even black officers may be biased in their enforcement.
It really seems like a tough nut to crack. I'm a mathematician, not a social scientist. I don't actually have any idea how to fix this.