LDA
The latent Dirichlet allocation model (LDA) is an unsupervised generative statistical topic model that explains why some parts of the data are similar \cite{ZHANG_2013}. The LDA result shows a number of 9 topics yields the best result \cite{Arun_2010}(Fig.\ref{159697}). To improve the efficiencies of our models, however, we decided to use a total of 5 topics which are also sufficient for our analysis. We calculated the weight of each topic for DC's and Marvel's reviews respectively, and the result (Fig.\ref{187161}) shows topic 4 is the best principal component for both DC and Marvel. However, Marvel has a slightly higher proportion(9.5%) of topic 4, and DC has a slightly higher proportion(8.1%) of topic 3. Through the word lists (Fig.\ref{107690}, \ref{383067}), we found that topic 4 represents the most common terms that people use when discussing superhero movies, while topic 3 is more related to the criticism and controversial feedback. Therefore, in comparison, Marvel's reviews are more positive and DC's are more controversial, which correspond to the previous results.