If it's the paper I'm thinking of then I read the original and that's not really a fair representation of what they did. What they actually did was they starte

If it's the paper I'm thinking of then I read the original and that's not really a fair representation of what they did. What they actually did was they started with some data on things which had gone viral but could only see how many were "infected" (number of views over time). They didn't have direct access to the granular information of who had seen what and who they shared or didn't share with. They decided to work backwards and model virality for some assumed network properties and contagiousness properties and then see which of their models had the similar statistics to actual viral posts. Network topology: iirc they tested the case of a gaussian distribution of connections meaning on average 95% of nodes had n connections +- 2sigma and they tested a power law type distribution meaning some thin minority of "influencers" had connections to nearly everyone". Contagious property: they tested the case where different content had a different virality property r0 with more or less contagious content all competing at the same time. They also tested the case of sheer randomness wherein all content is equally likely to be shared and which goes viral basically depends on the luck of who's eye it catches. The end result was that a power law and random contagiousness best fit the real world statistics.