STM

The Structural Topic Model (STM) is a general framework for topic modeling with document-level covariate information, which can improve inference and qualitative interpretability and are allowed to affect topical prevalence, topical content or both \cite{stm}. The topic number algorithm of Lee and Mimno (2014) shows a number of 90 topics yields the best result in STM model, but to reduce redundancy and to make the analysis easier, we set the number of topics as 5 same as the LDA model. As a result (Fig.\ref{201559}), we distinguish DC and Marvel with 2 obvious topics(1&4) while the other 3 topics remain neutral between DC and Marvel. Topic 1 that represents DC's reviews is featured by "bad" and "comic", and topic 4 that represents Marvel is featured by  "good" and "fun" (Fig.\ref{249695}). In conclusion, DC's reviews are more negative and Marvel's are more related to the fun and positive side. Such results are consistent with the LDA results.