Sadly, they don't really reveal any useful results, not even in the actual paper:
https://www.sciencedirect.com/science/article/pii/S266727742...
Apparently, "Business" is the category of what is being offered and it is the most important success factor (together with HSCS = social capital = creator popularity). But Figure 2 only lists 9 possible text keywords. And the publisher's page only says "Data will be made available on request."
Their boosted tree additionally reveals that projects from 2014 to 2016 were statistically more successful than average ... but that's not really actionable either. But that same table lists a high positive "Effect" for "journalism" on "Success", but journalism projects only have a 20% success rate with funding (and site-wide average is 40%).
At the end of the paper, they mention "three factors over which the crowdfunding founder has complete control: the goal, the number of reward options and the duration. To maximize the likelihood of success, Fig. 1 implies that all other things being equal, a founder should choose a Log(goal) which is less than 7, a number of reward options which is at least 40 and a duration of between 10 and 15 days."
"machine learning revealed that success actually increased up to about 750 comments, then leveled off."
It's easy to predict which projects are successful after the fact, but user comments is a feature which increments over the duration of a campaign, so you'd want to use a point in time count instead, which leads me to believe that this is not a skilled ML project.
did we need "machine learning" for this? is machine learning becoming the new term for basic statistics?