I think both ideas have merit. Which works best in practice would depend on the degree-of-shit (that's the scientific term) on your network. I fear special interest groups will, if they want, always outnumber legit researchers. If your network becomes large enough to attract the interest of these groups you might end up with "attacks" on climate research and other 'controversial' issues. Verified users could prevent this.
OTOH, I do not like excluding users, but I think public read-access and write-access trough accredited universities and research groups would be a fair balance. (Note, I also think if there is a charge associated with this process we should charge relative to the users country. Subsidizing access for developing countries by charging more in developed countries. To reduce exclusions.)
I don't see how special interest groups will be a problem. Surely, they can flood the reviews with unlegit attacks, but the academia can just ignore them and only stick to the high-reputation review groups. They can make a lot of noise, but I don't think it can be a big problem for the academia.
I'd say that from a UX experience you want the default-settings to work 99.9% of the time. This includes the ratings and reviews of articles. Which you could then only do by effectively censoring these groups when calculating ratings. If you're going to do that, why let them on in the first place?
OTOH, I do not like excluding users, but I think public read-access and write-access trough accredited universities and research groups would be a fair balance. (Note, I also think if there is a charge associated with this process we should charge relative to the users country. Subsidizing access for developing countries by charging more in developed countries. To reduce exclusions.)