“…Instead, we needed to rely on mixed-membership models (Airoldi, Blei, Fienberg, & Xing, 2008) where the assumption that a unit belongs to a single cluster is violated (Airoldi, Blei, Erosheva, & Fienberg, 2014). For the sake of identifying skill sets within job posts, we decided to adopt the mixed-membership model Latent Dirichlet Allocation, LDA (Blei, 2012), which has proven to work effectively at analyzing user-generated content like job posts (Ma, Zhang, Liu, Li, & Yuan, 2016).…”