Liying Wang 7/10/2020
library(tidyverse)
library(tidytext)
library(rvest)
library(data.table)
library(ggbeeswarm)
library(ggridges)
library(viridis)
library(purrr)Manga are Japanese comics or graphic novels that are very popular in the present day, especially in Asia. This is my exploration of the top 100 Manga since 1982 based on the data collected from https://www.anime-planet.com/manga/top-manga. The data includes the top 100 Manga, the year when they were first published, and the tags describing their topics.
# create base url for multi-page
url <- "https://www.anime-planet.com/manga/top-manga?page="
# download data from URL
info_manga <-
url %>%
map2_chr(1:3, paste0) %>%
map(. %>%
read_html() %>%
html_nodes("td.tableYear,
.tableTitle,
td.tableRank") %>%
html_text()) %>%
na_if("") %>%
unlist()# work on top 100
df_manga_top_100 <-
matrix(info_manga,
ncol = 3,
byrow = TRUE) %>%
as_data_frame() %>%
rename_at(vars(c('V1', 'V2', 'V3')),
~ c('rank', 'title', 'year')) %>%
# covert character to numeric
mutate(rank = as.numeric(rank),
# make subset for every rank 10
# covert numeric to character
rank_groups = as.factor(cut_interval(
1:nrow(.),
n = 4,
labels = FALSE))) %>%
# need to update the NA with correct year periodically
mutate(year = replace_na(year, "2019")) %>%
mutate(year = as.numeric(year),
period = case_when(`year` %in% 1989:2000 ~ "1989-2000",
`year` %in% 2001:2010 ~ "2001-2010",
`year` %in% 2011:2020 ~ "2011-2020",
TRUE ~ "other")) %>%
mutate(type = ifelse(grepl("(Light Novel)", title),
"light novel", "manga")) %>%
mutate(title = str_remove(title, "\\(Light Novel\\)")) %>%
mutate(year = as.factor(year),
type = as.factor(type)) %>%
# make sure they are in order by rank
arrange(rank) %>%
slice(1:100)Here are some steps to view the interactive plot:
- Click on this
launch binderbutton. to open Rstudio in the browser. It may take a few minutes.
- Click on README.Rmd in the lower right pane to open this file, which will show up at the upper left pane.
- Run the code chunks from the beginning and interact with the first plot by moving your mouse cursor over the point to see the title of each Manga.
library(plotly)
manga_top_100_inter <-
df_manga_top_100 %>%
pivot_wider(names_from = type,
values_from = rank)
top_100 <- plot_ly(manga_top_100_inter,
x = ~ manga,
y = ~ year,
name = "manga",
type = 'scatter',
mode = "markers",
text = ~title,
marker = list(color = "blue"),
hovertemplate = paste('rank: %{x}', '<br>%{text}<br>'),
texttemplate = '%{text}', textposition = 'outside')
top_100 <- top_100 %>%
add_trace(x = ~ `light novel`,
y = ~ year,
name = "light novel",
type = 'scatter',
mode = "markers",
text = ~title,
marker = list(color = "pink"),
hovertemplate = paste('rank: %{x}', '<br>%{text}<br>'),
texttemplate = '%{text}',
textposition = 'outside')
top_100_inter <-
top_100 %>%
layout(title = "Top 100 Manga and Light Novel",
xaxis = list(title = "rank"),
margin = list(l = 100))We can see that there are many popular Manga in the recent years after 2012, which shows that the newer the Manga, the more popular it is. However, three out of five of the top five Manga were released before 2002, such as Fullmetal Alchemist, Berserk, and One piece, representing the classic, especially Fullmetal Alchemist, also one of my favorites.
# plot the top 50 Manga by year
library(ggrepel)
df_manga_top_100 %>%
slice(1:50) %>%
ggplot(aes(year, rank)) +
geom_point(aes(color = type)) +
scale_colour_viridis_d(name = "type", direction = -1,
labels = c("light\nnovels", "comics")) +
geom_text_repel(aes(label = title, color = type), size = 2.5) +
theme_minimal() +
theme(legend.position = "right",
plot.title = element_text(hjust = 0.5)) +
labs(title = "Top 50 Manga from anime-planet.com") +
scale_y_reverse(limits = c(50, 1),
breaks = c(seq(50, 1,by = -10), 1)) +
theme(axis.text.x = element_text(angle = 45,
vjust = 0.5)) 
The bar plot shows that the higher-ranking Manga (top 1-25) seems to be distributed throughout the years with a higher number in 2016-2018. Despite the total number of Manga increasing with time after 2012, the proportion of lower-ranking ones (top 76-100) is also higher. In addition, we can see that Manga in 2015 and 2020 seems to have a lower rank in general according to the box plot.
# bar plot by count over time
ggplot(df_manga_top_100) +
geom_bar(aes(year, fill = rank_groups)) +
scale_y_continuous(limits = c(0,16), breaks = c(seq(0,16,by = 2), 16)) +
scale_fill_viridis_d(name = "Rank",
labels = c("1-25", "26-50", "51-75", "76-100")) +
theme_minimal() +
theme(axis.text.x = element_text(angle = 45,
vjust = 0.5)) 
# boxplot by counts based on ranking over time
df_manga_top_100 %>%
#mutate(year = fct_reorder(year, rank)) %>%
#remove above line to get order in year, .desc = TRUE can reverse the order
ggplot(aes(x=year, y=rank, fill=year)) +
geom_boxplot() +
geom_jitter(alpha=0.8, size=0.9) +
scale_fill_viridis(discrete=TRUE) +
scale_y_reverse(limits = c(100, 1),
breaks = c(seq(100, 1,by = -10), 1)) +
theme_minimal() +
theme(legend.position="none") +
labs(x="", y= "Rank from top 1 to 100") +
theme(axis.text.x = element_text(angle = 45,
vjust = 0.5)) 
# get page for each story
top_manga_urls <-
url %>%
map2_chr(1:3, paste0) %>%
map(. %>%
read_html() %>%
html_nodes("td a") %>%
html_attr('href')) %>%
unlist()
top_manga_urls <-
str_glue("https://www.anime-planet.com{top_manga_urls}")
# we will go to the page for each story and get the user stats
manga_story_stats <-
map(top_manga_urls[1:10],
~.x %>%
read_html() %>%
html_nodes(".status2 .slCount"))
# tidy the stats
map(manga_story_stats,
~.x %>%
html_text())
# go to the page for each story and get the tags
manga_story_tags <-
map(top_manga_urls,
~.x %>%
read_html() %>%
html_nodes(".tags a") %>%
html_text() %>%
tibble(text = .))
# convert the lists of tables to one big table
df <-
rbindlist(manga_story_tags, idcol = 'rank') %>%
mutate(rank = as.character(rank)) %>%
mutate(text = str_remove_all(text, "\\n"))Drama is the most popular genre, followed by action, fantasy, comedy, and romance. Among those genres, “Seinen” (https://en.wikipedia.org/wiki/Seinen_manga) is a type of Manga aimed at a younger audience especially men that could cover a wide range of topics. “BL” is also a special kind of genre that means Boys Love (https://en.wikipedia.org/wiki/Yaoi).
# combine manga top 100 info and tags
df_manga_t100_tags <-
matrix(info_manga,
ncol = 3,
byrow = TRUE) %>%
as_data_frame() %>%
rename_at(vars(c('V1', 'V2', 'V3')),
~ c('rank', 'title', 'year')) %>%
mutate(year = ifelse(year == "", NA, year)) %>%
mutate_all(any_vars(replace_na(.,"2018"))) %>%
left_join(df) %>%
mutate(text = ifelse(text == "Manhua", "Manhwa", text)) %>%
mutate(text = ifelse(text == "Shounen", "Seinen", text)) %>%
mutate(text = str_remove(text, ","))## Joining, by = "rank"
# ploting the most common tags from top 100 manga
tags_all <-
df_manga_t100_tags %>%
count(text, sort = TRUE)
tags_com <-
tags_all %>%
filter(n > 15) %>%
filter(!text %in% c("Manhwa", "Full Color",
"Webtoon", "Light Novel",
"Adapted to Anime")) %>%
mutate(text = reorder(text, n))
ggplot(tags_com,
aes(text, n)) +
geom_col() +
theme_minimal() +
coord_flip() +
labs(y = "frequency", x ="genre", title = "Top 100 Manga: Popular genre")
The box plot indicates that Seinen is the most common genre since 1982, followed by action. It also shows that BL and Manga based on a web novel appear and become popular after 2012.
# extract common tags to a list
list_com_tags <-
pull(tags_com, text)
# filter those common tags from the full dataset
rate_tags <-
df_manga_t100_tags %>%
mutate(rank = as.numeric(rank)) %>%
mutate(year = as.numeric(year)) %>%
filter(text %in% list_com_tags)
# box plot
ggplot(rate_tags,
aes(reorder(text, -year),
year)) +
geom_boxplot() +
#geom_quasirandom(alpha = 0.8) +
geom_beeswarm(alpha = 0.5) +
coord_flip() +
scale_y_continuous(limits = c(1980, 2020),
breaks = c(seq(1980, 2020, 2)),
name = "Year") +
theme_minimal() +
theme(axis.text.x = element_text(angle = 45,
vjust = 0.5)) +
labs(x ="", title = "Popular genre in Top 100 manga over time")
The histogram plots show those genres seem to increase significantly after 2012, especially the genre of drama, comedy, fantasy, and action.
# plot barplot for years by common tags
rate_tags %>%
ggplot() +
geom_bar(aes(year)) +
facet_wrap(~text,
ncol = 2) +
theme_minimal() +
scale_x_continuous(limits = c(1980, 2020),
breaks = c(seq(1980, 2020, 2)),
name = "Year") +
scale_y_continuous(limits = c(0, 8),
breaks = c(seq(0, 8, 2)),
name = "Manga per year") +
theme(axis.text.x = element_text(angle = 90,
vjust = 0.5)) 
library(scales)
# tidy the data to create rank groups for comparison
manga_t100_tags_groups <-
df_manga_t100_tags %>%
mutate(rank = as.numeric(rank), year = as.numeric(year)) %>%
mutate(rank_2groups = ifelse(rank < 51, "1-50", "51-100"),
year_2groups = ifelse(year < 2013, "before 2012", "after 2012"))In the plot, words on the right of the line are the genre that are found more after 2012, and words in red refer to a higher proportion in general. The words near the line are the genre found both in the early and later time with a similar proportion. The result shows that the topic increases and becomes diverse after 2012 with different focus, such as school life, historical, and martial arts.
# plot the comparison of genre frequency of Manga for different time periods
fre_year_groups <-
manga_t100_tags_groups %>%
count(year_2groups, text) %>%
mutate(proportion = n / sum(n)) %>%
select(-n) %>%
pivot_wider(names_from = year_2groups,
values_from = proportion)
ggplot(fre_year_groups,
aes(x = `after 2012`,
y = `before 2012`,
color = abs(`before 2012` - `after 2012`))) +
geom_abline(color = "gray40", lty = 2) +
geom_jitter(alpha = 0.5, size = 0.5, width = 0.1, height = 0.1) +
geom_text(aes(label = text), check_overlap = TRUE, vjust = 0.2) +
scale_x_log10(labels = percent_format()) +
scale_y_log10(labels = percent_format()) +
scale_color_gradient(limits = c(0, 0.01),
low = "lavenderblush4",
high = "red3") +
theme(legend.position = "none") +
labs(y = "before 2012",
x = "after 2012")