Great question @fada.emma!
Currently I am not sure how this works at CoinGossip. That is exactly the biggest challenge for current sentiment analytics companies like Clarabridge, Medallia and Cambridge Semantics, to algorithmically determine through machine learning as accurately as possible the sentiment relative to the rest. So it will be the same challenge for websites like CoinGossip.
As far I know, the machine learning capability is still at its infancy stage and a lot of manual stuff needs to be done. Think about classifying all words on a negative-positive scale given certain context (other words, subjects, country, language etc.).
Hopefully we'll see some relatively accurate tools in a year or two!
Boss, truly the learning capability of the machines is possibly in its babyhood which eventually places a 'question mark' on the integrity and authenticity of the manual forces behind the scenes.
Meanwhile, I join you in hoping for a better future 🔮.
Cheers @cryptotem