Question about umass coherence

[Gennaro Castaldo]

Hi everyone, sorry for the probably stupid question, but i can't seem to understando the centre of the problem. 
I am a novice to the topic modeling problem, and i was wondering about the meaning of the u_mass compared to the c_v. In other words, i seem to understand that for the c_v the highest coherence is the better, as it's common in this metrics. But for the u_mass implementation the negative values let me think that the matric has got it's maximum in 0 and values are generally negative. so the closest to zero ( from negative values), the better is the number of topics? Yet i fail to understand why i get the better results with low number of topics. 

Thanks for the help
Gennaro Castaldo