Some unsolicited advice: If anything, your consistent users have a lot of ways to discuss papers: classrooms, sections, MOOCs, stack* questions, conferences, proceedings, editorial pages of journals, invited lectures, ... In fact, a large of what people who read & write papers do is share and discuss them.

Another problem is, since your "social object" cannot maintain enough synchronous interaction, it becomes a dead social network. Not enough people will read & discuss the same paper at the same time. If you are going to fix one problem, this is it. # of papers or # registered users is not important, The only important metric is #time spent between me leaving a comment on the paper I'm currently reading & someone else liking, responding, etc. in a relevant way. Otherwise people have no reason to go there, it's an empty room.

So suppose you restrict the papers. Then it starts looking a lot like a scientific journal with social features - which a lot of sci journals are also doing, without a lot of interaction going on (see comments sections of PLoS journals).

As others commented - tough market. Try to elucidate a narrow user segment & journey using a model like "Crossing the Chasm".