As R analyses on Game of Thrones, Star Wars or The Lord of The Rings have already been done, I decided to visualize the superhero network of the nowadays very popular Marvel movies saga.
Scraping Marvel superheros
As often, the data can be found on Wikipedia. But surprisingly, the data needed was not avaiable in a English article, but within this French article.
Let’s scrape this wikitable with the rvest package.
library(tidyverse)
library(rvest)
# permanente link for reproductibility
url <- "https://fr.wikipedia.org/w/index.php?title=Liste_des_films_de_l%27univers_cin%C3%A9matographique_Marvel&oldid=144793972#Personnages"
marvel_df <- url %>%
read_html() %>%
html_nodes(".wikitable") %>%
html_table(fill = TRUE) %>%
.[[5]]Now that we have the data, we need to clean it as well as translate some names in English. We also have to distinguish the movie names from the characters.
marvel_df[1,1] <- "Iron Man 1"
marvel_df[2,1] <- "The Incredible Hulk"
marvel_df[4,1] <- "Thor 1"
marvel_df[8,1] <- "Thor: The Dark World"
marvel_df[9,1] <- "Captain America 2"
marvel_df[10,1] <- "Guardians of the Galaxy"
marvel_df[11,1] <- "Avengers: Age of Ultron"
marvel_df[12,1] <- "Ant-Man 1"
marvel_df[14,1] <- "Doctor Strange 1"
marvel_df[15,1] <- "Guardians of the Galaxy Vol. 2"
marvel_df[18,1] <- "Black Panther 1"
marvel_df[20,1] <- "Ant-Man and the Wasp"
marvel_df[21,1] <- "Captain Marvel 1"
marvel_df <- marvel_df %>%
rename("Black Widow" = "Veuve noire",
"Hawkeye" = "Œil de Faucon",
"Scarlet Witch" = "Sorcière rouge") %>%
mutate(Film = factor(Film, levels = unique(Film))) %>%
mutate_all(funs(str_replace_all(., c("Oui" = "1", "^$" = "0")))) #^$ is for empty stringReproducing the wikitable with a heatmap
Let’s tidy the data in order to reproduce the Wikipedia table with ggplot2. Note that the characters and movies are in a different order.
marvel_tidy <- marvel_df %>%
reshape2::melt(id.vars = "Film", value.name = "Value") %>%
rename("Character" = "variable")
ggplot(marvel_tidy, aes(x = Character, y = Film)) +
geom_tile(aes(fill = Value)) +
scale_fill_manual(values=c("0"="grey", "1"="lightgreen"),
name="", labels=c("Out","In")) +
theme_minimal() +
theme(plot.title = element_text(face = "bold", hjust = 0.5),
plot.caption = element_text(colour = "dimgrey"),
axis.text.x = element_text(angle = 45, vjust = 1, hjust=1)) +
labs(title = "Characters appearance in the Marvel Movies",
caption = "Félix Luginbühl (@lgnbhl)\nData source: Wikipedia")
Centrality metrics
In order to know which characters and movies are the most central, we can use indicators like the degree (number of ties) and the closeness (centrality based on distance to others in the graph).
library(tidygraph)
marvel_graph <- marvel_tidy %>%
filter(Value != "0") %>%
select(-Value) %>%
as_tbl_graph(directed = FALSE) %>%
mutate(degree = centrality_degree(),
closeness = centrality_closeness(),
betweenness = centrality_betweenness()) %>%
#create type variable
full_join(tibble(film = marvel_df$Film, type = "Movie"), by = c("name" = "film")) %>%
mutate(type = replace_na(type, "Character"))
marvel_graph %>%
activate(nodes) %>%
as_tibble() %>%
arrange(desc(degree))## # A tibble: 36 x 5
## name degree closeness betweenness type
## <chr> <dbl> <dbl> <dbl> <chr>
## 1 Avengers: Infinity War 14. 0.0161 237. Movie
## 2 Captain America: Civil War 10. 0.0128 57.8 Movie
## 3 Avengers: Age of Ultron 9. 0.0135 61.6 Movie
## 4 Iron Man 9. 0.0135 89.2 Character
## 5 Captain America 9. 0.0135 69.3 Character
## 6 Nick Fury 8. 0.0111 94.0 Character
## 7 Avengers 7. 0.0128 40.7 Movie
## 8 Ant-Man 7. 0.0123 35.3 Movie
## 9 Thor 7. 0.0128 62.2 Character
## 10 Hulk 6. 0.0125 40.1 Character
## # ... with 26 more rowsThe central characters are Iron Man and Captain America (ex aequo), followed by Nick Fury, Thor and Hulk. The central movies are the two Avengers movies and as well as Captain America: Civil War (2016).
Who is the most distant character from Iron Man?
marvel_graph %>%
mutate(distance = bfs_dist(name == "Iron Man", mode = "all")) %>%
filter(type == "Character") %>%
select(name, distance) %>%
arrange(desc(distance))## # A tbl_graph: 15 nodes and 0 edges
## #
## # An undirected simple graph with 15 components
## #
## # Node Data: 15 x 2 (active)
## name distance
## <chr> <int>
## 1 Captain Marvel 4
## 2 Captain America 2
## 3 Nick Fury 2
## 4 Thor 2
## 5 Hulk 2
## 6 Black Widow 2
## # ... with 9 more rows
## #
## # Edge Data: 0 x 2
## # ... with 2 variables: from <int>, to <int>Captain Marvel is the most distant character. However, as the movie introducing Captain Marvel will only be released in 2019, other Marvel characters could be added to the movie (and therefore reducing the distance from Iron Man).
Plotting the network
Now let’s visualize the centrality degree of the movies and characters using {ggraph}.
library(ggraph)
set.seed(100)
ggraph(marvel_graph, layout = "nicely") +
geom_edge_diagonal(alpha = 0.2) +
geom_node_point(aes(size = degree, color = as.factor(type)), alpha = 0.8) +
scale_color_brewer(palette = "Set1", name = "Type") +
geom_node_text(aes(label = name), size = 2.5, repel = TRUE) +
theme_graph() +
theme(plot.background = element_rect(fill = "#f8f2e4")) +
labs(title = "Centrality in the Marvel Cinematic Universe",
size = "Degree",
caption = "Félix Luginbühl (@lgnbhl)\n Data source: Wikipedia")
Now let’s try some clustering, with the Walktrap algorithm.
set.seed(100)
marvel_graph %>%
activate(nodes) %>%
mutate(group_walktrap = group_walktrap()) %>%
ggraph(layout = "nicely") +
geom_edge_diagonal(alpha = 0.2) +
geom_node_point(aes(color = as.factor(group_walktrap), shape = as.factor(type))) +
geom_node_text(aes(label = name), size = 2.5, alpha = 0.8, repel = TRUE) +
scale_color_brewer(palette = "Set1", name = "Walktrap Group") +
theme_graph() +
theme(plot.background = element_rect(fill = "#f8f2e4")) +
labs(title = "Clustering the Marvel Cinematic Universe",
shape = "Type",
caption = "Félix Luginbühl (@lgnbhl)\n Data source: Wikipedia")
The walktrap algorithm is doing a good job, as the characters and the movies seem correctly grouped.
An interactive visualization of the Marvel Universe Network
Making an interactive network with {visNetwork} is quite easy. Just play with the interactive network below.
library(visNetwork)