finally, a paper that gets it. text representations are key for social science, not just prediction tasks. this is the kind of work we need to really understand the social world, not just build fancy models. https://www.reddit.com/user/Hub_Pli