R-Package to retrieve web data from the Internet Archive
The goal of the archiveRetriever package is to provide a systematic workflow for retrieving web data from mementos stored in the Internet Archive. Currently, the package includes the following functions:
archive_overviewgenerates a calendar providing an overview of the available mementos of the homepage in the Internet Archive within a specific time range. This function is very useful for getting a quick glimpse of the web data available when planning to retrieve a comprehensive coverage of the homepage from the Internet Archiveretrieve_urlsgenerates a vector of links to mementos of the homepage stored in the Internet Archive within a specific time range.retrieve_linksgenerates a tibble with two columns including the link to the memento of the homepage stored in the Internet Archive as well as all links within the memento. The two column translate to the parent link with its child references. This function is useful to fully cover the content within a homepage for retrieval.scrape_urlsgenerates a tibble including the link of the memento being scraped as well as the scraped content structured in different columns. The number of columns for the scraped content amounts to the length of the XPath or CSS selectors used to scrape the content.
We present a short step-by-step guide as well as the functions in more detail below.
A stable version of archiveRetriever can be directly accessed on CRAN:
install.packages("archiveRetriever", force = TRUE)To install the latest development version of archiveRetriever directly
from GitHub use:
library(devtools) # Tools to Make Developing R Packages Easier
devtools::install_github("liserman/archiveRetriever")First, load the package
library(archiveRetriever) # Systematically retrieving web data from the Internet ArchiveIn the following, we are going to exemplify the workflow of the package using the mementos of the New York Times online version stored in the Internet Archive.
The workflow of the package follows a simple rule:
-
Get an overview of data availability in the Internet Archive
-
Retrieve the mementos of the base url from the Internet Archive
-
Retrieve the links within the base url from the memento stored in the Internet Archive (only necessary when scraping complete homepages)
-
Scrape the content and get it conveniently stored in tibbles.
As the Internet Archive is not able to archive the complete internet it is always important to check whether the memento of the homepage you want to scrape is actually available in the Internet Archive.
nytimes_overview <- archive_overview(homepage = "https://www.nytimes.com/",
startDate = "2020-10-01",
endDate = "2020-12-31")The archive_overview function creates a calendar providing an overview
of the homepage’s availability in the Internet Archive.
nytimes_overview
For the New York Times, we find that the Internet Archive save a memento of their homepage every day, which is highly reasonable as this homepage is one of the most visited homepages on the internet.
Next to base urls, the Internet Archive also stores child urls as
mementos. Using the archive_overview function, it is of course also
possible to get a calendar showing the availability of mementos of
specific child urls (for example the article of the New York Times on
the election of Joe Biden as 46. President of the USA).
nytimesArticle_overview <- archive_overview(homepage = "https://www.nytimes.com/2020/11/07/us/politics/biden-election.html",
startDate = "2020-10-01",
endDate = "2020-12-31")nytimesArticle_overview
As the article has been published on November 07, there are of course no mementos available before that date.
The Internet Archive stores mementos of homepages in their archive which allows researchers to retrieve historical content from the internet or examine changes to existing homepages. Scraping content from the Internet Archive often requires retrieving mementos from a certain time range or specific points in time.
Applying the retrieve_urls function on a homepage results in a
character vector of mementos of the homepage available from the Internet
Archive.
nytimes_mementos <- retrieve_urls(homepage = "https://www.nytimes.com/",
startDate = "2020-10-01",
endDate = "2020-12-31")In the Internet Archive often more than one memento is stored each day.
For convenience, the retrieve_urls only retrieves one memento for each
day.
nytimes_mementos[1:10]
#> [1] "http://web.archive.org/web/20201001000041/https://www.nytimes.com/"
#> [2] "http://web.archive.org/web/20201002000016/http://nytimes.com/"
#> [3] "http://web.archive.org/web/20201003000006/https://nytimes.com/"
#> [4] "http://web.archive.org/web/20201004000201/https://www.nytimes.com/"
#> [5] "http://web.archive.org/web/20201005000047/http://nytimes.com/"
#> [6] "http://web.archive.org/web/20201006000036/http://nytimes.com/"
#> [7] "http://web.archive.org/web/20201007000202/https://www.nytimes.com/"
#> [8] "http://web.archive.org/web/20201008000222/https://www.nytimes.com/"
#> [9] "http://web.archive.org/web/20201009000201/https://www.nytimes.com/"
#> [10] "http://web.archive.org/web/20201010000605/http://nytimes.com/"For many scraping applications, researchers aim to extract information from all links within a homepage to get a complete picture of the information stored, e.g. when scraping news content from online newspapers, blogs on Reddit or press releases published by political parties.
Sticking to the example of the New York Times, we extract all links of
the memento stored on October 01, 2020 using the retrieve_links
function. Please be aware that the retrieve_links function only takes
mementos of the Internet Archive as input to ensure only these pages are
being scraped using our scraping functions.
nytimes_links <- retrieve_links(ArchiveUrls = "http://web.archive.org/web/20201001000041/https://www.nytimes.com/")The retrieve_links function results in a tibble with two columns,
including the base url of the memento in the first column and all links
in the second column. From this, user of this function might decide to
filter out links which do not point to content relevant for analysis
using packages for string operations, such as
stringr.
head(nytimes_links)
#> # A tibble: 6 x 2
#> baseUrl links
#> <chr> <chr>
#> 1 http://web.archive.org/web/20201001000041/https://www.nytimes.com/ http://web~
#> 2 http://web.archive.org/web/20201001000041/https://www.nytimes.com/ http://web~
#> 3 http://web.archive.org/web/20201001000041/https://www.nytimes.com/ http://web~
#> 4 http://web.archive.org/web/20201001000041/https://www.nytimes.com/ http://web~
#> 5 http://web.archive.org/web/20201001000041/https://www.nytimes.com/ http://web~
#> 6 http://web.archive.org/web/20201001000041/https://www.nytimes.com/ http://web~Sometimes, some of the retrieved urls are unable to access, producing an
error message. Especially when retrieving links from a larger body of
archive-urls, you may not want your process to break due to these
inaccessible urls. For this purpose, you can use the ignoreErrors
option.
nytimes_links <- retrieve_links(ArchiveUrls = "http://web.archive.org/web/20201001000041/https://www.nytimes.com/", ignoreErrors = TRUE)For some applications, it might not be necessary to include the
retrieve_links function into the workflow. When only interested in one
specific homepage, it can be sufficient to only retrieve the mementos
using the retrieve_urls function.
The scrape_urls function is the main function of the
ArchiveRetriever package. The function takes a memento of the Internet
Archive and a XPath (or CSS) vector as obligatory inputs and results in
a tibble with the content scraped using the XPath/CSS selectors. There
is one important point to consider when entering the Paths for
scraping: The option only takes named vectors, in order to provide
meaningful column names for the resulting tibbles.
nytimes_article <- scrape_urls(Urls = "http://web.archive.org/web/20201001004918/https://www.nytimes.com/2020/09/30/opinion/biden-trump-2020-debate.html",
Paths = c(title = "//h1[@itemprop='headline']",
author = "//span[@itemprop='name']",
date = "//time//text()",
article = "//section[@itemprop='articleBody']//p"))nytimes_article
#> # A tibble: 1 x 5
#> Urls title author date article
#> <chr> <chr> <chr> <chr> <chr>
#> 1 http://web.archive.o~ After That Fiasc~ Frank B~ Sept.~ "I wasn’t in the crow~When using the scrape_urls function to scrape large amounts of urls,
we added some important (optional) features, ensuring that the scraping
process works smoothly. Most importantly, the process breaks when no
content could be scraped for a certain number of urls (default is 10) -
most often meaning that the XPath have not been selected correctly.
Additionally, the process breaks when only some elements of the Paths
could be scraped - implying that the XPaths have changed for parts of
the content aimed to be scraped. After break-off, the function still
outputs a tibble, ensuring that the process does not need to be started
anew. After break-off, users need to identify the element position of
break-off, fix the error in the Paths and are able to re-start the
scrape_urls function exactly where it broke off. The scraping process
until break-off can be attached to the new process via the attachto
option. Lastly, we also added a length warning for long url vector
inserted to the scrape_urls raising awareness that a large chunk of
data is now going to be scraped.
All these different break-offs can be deactivated using the optional
arguments ignoreErrors = TRUE and stopatempty = FALSE. However, we
advise to use these options with care. The cutoff point for the number
of urls without content in a row until break-off can be set using
emptylim. In order to facilitate the automation of the scraping
process, we also added the optional argument lengthwarning = FALSE to
deactivate the length warning.
In many scraping projects it is very convenient to collapse similar
html-nodes into a single observation. We have also done this in the
example outlined so far. This enabled us to retrieve the clean article
text, filtering out all figures, advertisements, teasers and other
annoying things newspapers put in between paragraphs. However, for some
applications, we may want to treat similar html-nodes as independent
observations. One example for this is the retrieving of article
headlines and teasers from overview pages. For this purpose,
scrape_urls offers the option collapse = FALSE. Instead of
collapsing the content of similar html-nodes into one observation,
different nodes are treated as independent observations. Yet, in order
for this to work as intended, it is necessary that for all different
Paths you extract in the same process, the number of retrieved nodes
matches.
nytimes_teaser <- scrape_urls(Urls = "https://web.archive.org/web/20201001000859/https://www.nytimes.com/section/politics",
Paths = c(title = "//article/div/h2//text()",
teaser = "//article/div/p/text()"),
collapse = FALSE,
archiveDate = TRUE)nytimes_teaser
#> # A tibble: 4 x 4
#> Urls title teaser archiveDate
#> <chr> <chr> <chr> <date>
#> 1 https://web.archive.o~ Tuesday’s Debate Mad~ President Trump’s un~ 2020-10-01
#> 2 https://web.archive.o~ Takeaways on Trump, ~ A New York Times Mag~ 2020-10-01
#> 3 https://web.archive.o~ Biden Tours Ohio and~ A day after a turbul~ 2020-10-01
#> 4 https://web.archive.o~ Six Takeaways From t~ It was 90 minutes of~ 2020-10-01Lastly, scrape_urls comes with a handy option archiveDate, to add a
column indicating the date of the retrieved Internet Archive memento.
All these different break-offs can be deactivated using the optional
arguments ignoreErrors = TRUE and stopatempty = FALSE. However, we
advise to use these options with care. The cutoff point for the number
of urls without content in a row until break-off can be set using
emptylim. In order to facilitate the automation of the scraping
process, we also added the optional argument lengthwarning = FALSE to
deactivate the length warning.
archivedate collapse