57 Lab: University of Edinburgh Art Collection
The University of Edinburgh Art Collection “supports the world-leading research and teaching that happens within the University. Comprised of an astonishing range of objects and ideas spanning two millennia and a multitude of artistic forms, the collection reflects not only the long and rich trajectory of the University, but also major national and international shifts in art history.”3.
Note: See the sidebar here and note that there are 3294 pieces in the art collection we’re collecting data on.
In this lab, we’ll scrape data on all art pieces in the Edinburgh College of Art collection.
The learning goals of this lab are:
- Web scraping from a single page
- Writing functions
- Iteration
- Writing data
Before getting started, let’s check that a bot has permissions to access pages on this domain.
##
collections.ed.ac.uk
## [1] TRUE
Getting started
Go to the course GitHub organization and locate your lab repo, which should be named lab-08-uoe-art
. Grab the URL of the repo, and make a template for your own copy.
R scripts vs. R Markdown documents
Today we will be using both R scripts and R Markdown documents:
.R
: R scripts are plain text files containing only code and brief comments,- We’ll use R scripts in the web scraping stage and ultimately save the scraped data as a csv.
.Rmd
: R Markdown documents are plain text files containing.- We’ll use an R Markdown document in the web analysis stage, where we start off by reading in the csv file we wrote out in the scraping stage.
Here is the organization of your repo, and the corresponding section in the lab that each file will be used for:
|-data
| |- README.md
|-lab-08-uoe-art.Rmd # analysis
|-lab-08-uoe-art.Rproj # project management
|-README.md
|-scripts # webscraping
| |- 01-scrape-page-one.R # scraping a single page
| |- 02-scrape-page-function.R # functions
| |- 03-scrape-page-many.R # iteration
SelectorGadget
For this lab, I recommend using Google Chrome as your web browser. In case you haven’t installed the SelectorGadget extension… go to the SelectorGadget extension page on the Chrome Web Store and click on “Add to Chrome” (big blue button). A pop up window will ask Add “SelectorGadget”?, click “Add extension”.
Another pop up window will ask whether you want to get your extensions on all your computer. If you want this, you can turn on sync, but you don’t need to for the purpose of this lab.
You should now be able to access SelectorGadget by clicking on the icon next to the search bar in the Chrome browser.
Scraping a single page
Tip: To run the code you can highlight or put your cursor next to the lines of code you want to run and hit Command+Enter.
Work in scripts/01-scrape-page-one.R
.
We will start off by scraping data on the first 10 pieces in the collection from here.
First, we define a new object called first_url
, which is the link above. Then, we read the page at this url with the read_html()
function from the rvest package. The code for this is already provided in 01-scrape-page-one.R
.
# set url
first_url <- "https://collections.ed.ac.uk/art/search/*:*/Collection:%22edinburgh+college+of+art%7C%7C%7CEdinburgh+College+of+Art%22?offset=0"
# read html page
page <- read_html(first_url)
For the ten pieces on this page, we will extract title
, artist
, and link
information, and put these three variables in a data frame.
57.0.1 Titles
Let’s start with titles. We make use of the SelectorGadget to identify the tags for the relevant nodes:
## {xml_nodeset (10)}
## [1] <a href="./record/99801?highlight=*:*">Untitled ...
## [2] <a href="./record/21172?highlight=*:*">Portrait of a Seated Woman ...
## [3] <a href="./record/21146?highlight=*:*">Seated Male Nude ...
## [4] <a href="./record/22658?highlight=*:*">Summer Fayre ...
## [5] <a href="./record/22534?highlight=*:*">Tomb ...
## [6] <a href="./record/21881?highlight=*:*">Untitled ...
## [7] <a href="./record/22703?highlight=*:*">Portrait Study ...
## [8] <a href="./record/53738?highlight=*:*">Vase ...
## [9] <a href="./record/122843?highlight=*:*">Composition with Two Women ...
## [10] <a href="./record/20809?highlight=*:*">Standing Male Nude ...
Then we extract the text with html_text()
:
## [1] "Untitled (1983)"
## [2] "Portrait of a Seated Woman (1965-1966)"
## [3] "Seated Male Nude (1959)"
## [4] "Summer Fayre "
## [5] "Tomb "
## [6] "Untitled (Unknown)"
## [7] "Portrait Study (1950)"
## [8] "Vase "
## [9] "Composition with Two Women (Circa 1972)"
## [10] "Standing Male Nude (1952)"
And get rid of all the spurious whitespace in the text with str_squish()
:
Take a look at the help docs for `str_squish()` (with `?str_squish`) to
## [1] "Untitled (1983)"
## [2] "Portrait of a Seated Woman (1965-1966)"
## [3] "Seated Male Nude (1959)"
## [4] "Summer Fayre"
## [5] "Tomb"
## [6] "Untitled (Unknown)"
## [7] "Portrait Study (1950)"
## [8] "Vase"
## [9] "Composition with Two Women (Circa 1972)"
## [10] "Standing Male Nude (1952)"
And finally save the resulting data as a vector of length 10:
Links
The same nodes that contain the text for the titles also contains information on the links to individual art piece pages for each title. We can extract this information using a new function from the rvest package, html_attr()
, which
extracts attributes.
A mini HTML lesson! The following is how we define hyperlinked text in HTML:
<a href="https://www.google.com">Seach on Google</a>
And this is how the text would look like on a webpage: Seach on Google.
Here the text is Seach on Google
and the href
attribute contains the url
of the website you’d go to if you click on the hyperlinked text: https://www.google.com
.
The moral of the story is: the link is stored in the href
attribute.
page %>%
html_nodes(".iteminfo") %>% # same nodes
html_node("h3 a") %>% # as before
html_attr("href") # but get href attribute instead of text
## [1] "./record/99801?highlight=*:*" "./record/21172?highlight=*:*"
## [3] "./record/21146?highlight=*:*" "./record/22658?highlight=*:*"
## [5] "./record/22534?highlight=*:*" "./record/21881?highlight=*:*"
## [7] "./record/22703?highlight=*:*" "./record/53738?highlight=*:*"
## [9] "./record/122843?highlight=*:*" "./record/20809?highlight=*:*"
These don’t really look like urls as we know them though. They’re relative links.
Note: See the help for
str_replace()
to find out how it works. Remember that the first argument is passed in from the pipeline, so you just need to define thepattern
andreplacement
arguments.
- Click on one of art piece titles in your browser and take note of the URL of the webpage it takes you to. How does that URL compare to what we scraped above? How is it different? Using
str_replace()
, fix the URLs.
Functions
Work in scripts/02-scrape-page-function.R
.
You’ve been using R functions, now it’s time to write your own!
Let’s start simple. Here is a function that takes in an argument x
, and adds 2 to it.
Let’s test it:
## [1] 5
## [1] 12
The skeleton for defining functions in R is as follows:
Then, a function for scraping a page should look something like:
Tip: Function names should be short but evocative verbs.
function_name <- function(url) {
# read page at url
# extract title, link, artist info for n pieces on page
# return a n x 3 tibble
}
Fill in the blanks using code you already developed in the previous exercises. Name the function
scrape_page
.Test out your new function by running the following in the console. Does the output look right?
Iteration
Work in scripts/03-scrape-page-many.R
.
We went from manually scraping individual pages to writing a function to do the same. Next, we will work on making our workflow a little more efficient by using R to iterate over all pages that contain information on the art collection.
Reminder: The collection has 3294 pieces in total, as of the last time this page was compiled.
That means we develop a list of URLs (of pages that each have 10 art pieces), and write some code that applies the scrape_page()
function to each page, and combines the resulting data frames from each page into a single data frame with 3294 rows and 3 columns.
List of URLs
Click through the first few of the pages in the art collection and observe their URLs to confirm the following pattern:
[sometext]offset=0 # Pieces 1-10
[sometext]offset=10 # Pieces 11-20
[sometext]offset=20 # Pieces 21-30
[sometext]offset=30 # Pieces 31-40
...
[sometext]offset=2900 # Pieces 2900-2909
We can construct these URLs in R by pasting together two pieces: (1) a common (root
) text for the beginning of the URL, and (2) numbers starting at 0, increasing by 10, all the way up to 2900. Two new functions are helpful for accomplishing this: paste0()
for pasting two pieces of text and seq()
for generating a sequence of numbers.
- Fill in the blanks to construct the list of URLs.
Mapping
Finally, we’re ready to iterate over the list of URLs we constructed. We will do this by mapping the function we developed over the list of URLs. There are a series of mapping functions in R (more details on that soon!), and they each take the following form:
map([x], [function to apply to each element of x])
In our case, x
is the list of URLs we constructed and the function to apply to each element of x
is the function we developed earlier, scrape_page
. Now, we want a data frame, so we use map_dfr
function:
- Fill in the blanks to scrape all pages, and to create a new data frame called
uoe_art
.
Analysis
Work in lab-08-uoe-art.Rmd
for the rest of the lab.
Now that we have a tidy dataset that we can analyze, let’s do that!
We’ll start with some data cleaning, to clean up the dates that appear at the end of some title text in parentheses. Some of these are years, others are more specific dates, some art pieces have no date information whatsoever, and others have some non-date information in parentheses. This should be interesting to clean up!
First thing we’ll try is to separate the title
column into two: one for the actual title
and the other for the date
if it exists. In human speak, we need to
“separate the
title
column at the first occurence of(
and put the contents on one side of the(
into a column calledtitle
and the contents on the other side into a column calleddate
”
Luckily, there’s a function that does just this: separate()
!
Once we have completed separating the single title
column into title
and date
, we need to do further cleanup in the date
column to get rid of extraneous )
s with str_remove()
, capture year information, and save the data as a numeric variable.
Hint: Remember escaping special characters from that video? Which video… oh you know the one. You’ll need to use that trick again.
Fill in the blanks to implement the data wrangling we described above. Note that this approach will result in some warnings when you run the code, and that’s OK! Read the warnings, and explain what they mean, and why we are ok with leaving them in given that our objective is to just capture
year
where it’s convenient to do so.Print out a summary of the data frame using the
skim()
function. How many pieces have artist info missing? How many have year info missing?Make a histogram of years. Use a reasonable bin width. Do you see anything out of the ordinary?
Find which piece has the out-of-the-ordinary year and go to its page on the art collection website to find the correct year for it. Can you tell why our code didn’t capture the correct year information? Correct the error in the data frame and visualize the data again.
Hint: You’ll want to use
mutate()
andif_else()
orcase_when()
to implement the correction.
Who is the most commonly featured artist in the collection? Do you know them? Any guess as to why the university has so many pieces from them?
Final question! How many art pieces have the word “child” in their title? See if you can figure it out, and ask for help if not.
Hint: yOU can use a combination of
filter()
andstr_detect()
. You will want to read the help forstr_detect()
at a minimum, and consider how you might capture titles where the word appears as “child” and “Child”.