Figure 3

Time evolution of word topics (left column) and legislation topics (right column). For each year y, we compute the topic distribution \(P(\boldsymbol{T}|y)\) (Section 2). (A-B) Time evolution of the importance of some illustrative topics in the housing corpus, normalized by their maximum importance across years (Section 2). While some topics decrease in the period 2012-2018, others show the opposite behavior. Topics depicted in blue reach their maximum importance in early years, whereas orange topics reach their maximum importance in the final years. Topics that reach their maximum importance in intermediate years are depicted in gray. (C-H) Evolution of all topics for the three corpora: (C-D) housing; (E-F) homicides, and (G-H) and condominium. Each layer corresponds to a topic and their thickness corresponds to their relative importance in a given year. Topics shown in panels (A) and (B) are identified in (C) and (D), respectively. Hierarchical levels: C, L = 2; D, L = 1; E, L = 2; F, L = 0; G, L = 1; H, L = 0; we show the analogous results for other hierarchical levels in the models in Figs. S3 and S4 in Additional file 1