That's fine if you're using it to make business decisions or embedded in a chatbot, but if you're using your new NLP algorithm to "discover" latent social biases and then make commentary .... then a "good amount of poking and tuning" is, at best, academic fraud.