Parallel Dynamic Web-Scraping Using 'RSelenium'.

parsel

parsel is a framework for parallelized dynamic web-scraping using RSelenium. Leveraging parallel processing, it allows you to run any RSelenium web-scraping routine on multiple browser instances simultaneously, thus greatly increasing the efficiency of your scraping. parsel utilizes chunked input processing as well as error catching and logging, to ensure seamless execution of your scraping routine and minimal data loss, even in the presence of unforeseen RSelenium errors. parsel additionally provides convenient wrapper functions around RSelenium methods, that allow you to quickly generate safe scraping code with minimal coding on your end.

Installation

# Install parsel from CRAN
install.packages("parsel")

# Or the development version from GitHub:
# install.packages("devtools")
devtools::install_github("till-tietz/parsel")

Usage

Parallel Scraping

The following example will hopefully serve to illustrate the functionality and ideas behind how parsel operates. We’ll set up the following scraping job:

1. navigate to a random Wikipedia article
2. retrieve its title
3. navigate to the first linked page on the article
4. retrieve the linked page’s title and first section

and parallelize it with parsel.

parsel requires two things:

1. a scraping function defining the actions to be executed in each RSelenium instance. Actions to be executed in each browser instance should be written in the conventional RSelenium syntax with remDr$ specifying the remote driver.
2. some input x to those actions (e.g. search terms to be entered in search boxes or links to navigate to etc.)

library(RSelenium)
library(parsel)

#let's define our scraping function input 
#we want to run our function 4 times and we want it to start on the wikipedia main page each time 
input <- rep("https://de.wikipedia.org",4)

#let's define our scraping function 

get_wiki_text <- function(x){
  input_i <- x
  
  #navigate to input page (i.e wikipedia)
  remDr$navigate(input_i)
  
  #find and click random article 
  rand_art <- remDr$findElement(using = "id", "n-randompage")$clickElement()
  
  #get random article title 
  title <- remDr$findElement(using = "id", "firstHeading")$getElementText()[[1]]
  
  #check if there is a linked page
  link_exists <- try(remDr$findElement(using = "xpath", "/html/body/div[3]/div[3]/div[5]/div[1]/p[1]/a[1]"))
  
  #if no linked page fill output with NA
  if(is(link_exists,"try-error")){
    first_link_title <- NA
    first_link_text <- NA
    
    #if there is a linked page
  } else {
    #click on link
    link <- remDr$findElement(using = "xpath", "/html/body/div[3]/div[3]/div[5]/div[1]/p[1]/a[1]")$clickElement()
    
    #get link page title
    first_link_title <- try(remDr$findElement(using = "id", "firstHeading"))
    if(is(first_link_title,"try-error")){
      first_link_title <- NA
    }else{
      first_link_title <- first_link_title$getElementText()[[1]]
    }
    
    #get 1st section of link page
    first_link_text <- try(remDr$findElement(using = "xpath", "/html/body/div[3]/div[3]/div[5]/div[1]/p[1]"))
    if(is(first_link_text,"try-error")){
      first_link_text <- NA
    }else{
      first_link_text <- first_link_text$getElementText()[[1]]
    }
  }
  out <- data.frame("random_article" = title,
                    "first_link_title" = first_link_title,
                    "first_link_text" = first_link_text)
  return(out)
}

Now that we have our scrape function and input we can parallelize the execution of the function. For speed and efficiency reasons, it is advisable to specify the headless browser option in the extraCapabilities argument. parscrape will show a progress bar, as well as elapsed and estimated remaining time so you can keep track of scraping progress.

wiki_text <- parsel::parscrape(scrape_fun = get_wiki_text,
                               scrape_input = input,
                               cores = 2,
                               packages = c("RSelenium","XML"),
                               browser = "firefox",
                               scrape_tries = 1,
                               extraCapabilities = list(
                                     "moz:firefoxOptions" = list(args = list('--headless'))
                                     ))

parscrape returns a list with two elements:

1. a list of your scrape function output
2. a data.frame of inputs it was unable to scrape, and the associated error messages

RSelenium Constructors

parsel allows you to generate safe scraping code with minimal hassle by simply composing constructor functions that effectively act as wrappers around RSelenium methods in a pipe. You can return a scraper function defined by constructors to the environment by starting your pipe with start_scraper() and ending it with build_scraper(). Alternatively you can dump the code generated by your constructor pipe to the console via show(). We’ll reproduce a slightly stripped down version of the RSelenium code in the above wikipedia scraping routine via the parselconstructor functions.

library(parsel)

# returning a scaper function 
start_scraper(args = "x", name = "get_wiki_text") %>>%
  click(using = "id", value = "'n-randompage'", name = "rand_art") %>>%
  get_element(using = "id", value = "'firstHeading'", name = "title") %>>%
  click(using = "xpath", value = "'/html/body/div[3]/div[3]/div[5]/div[1]/p[1]/a[1]'", name = "link") %>>%
  get_element(using = "id", value = "'firstHeading'", name = "first_link_title") %>>%
  get_element(using = "xpath", value = "'/html/body/div[3]/div[3]/div[5]/div[1]/p[1]'", name = "first_link_text") %>>%
  build_scraper()
#> [1] "scraping function get_wiki_text constructed and in environment"
#> [1] "scraping function get_wiki_text constructed and in environment"

ls()  
#> [1] "get_wiki_text"

# dumping generated code to console 
go(url = "x") %>>%
  click(using = "id", value = "'n-randompage'", name = "rand_art") %>>%
  get_element(using = "id", value = "'firstHeading'", name = "title") %>>%
  click(using = "xpath", value = "'/html/body/div[3]/div[3]/div[5]/div[1]/p[1]/a[1]'", name = "link") %>>%
  get_element(using = "id", value = "'firstHeading'", name = "first_link_title") %>>%
  get_element(using = "xpath", value = "'/html/body/div[3]/div[3]/div[5]/div[1]/p[1]'", name = "first_link_text") %>>%
  show()
#> # navigate to url
#> not_loaded <- TRUE
#> remDr$navigate(x)
#> while(not_loaded){
#> Sys.sleep(0.25)
#> current <- remDr$getCurrentUrl()[[1]]
#> if(current == x){
#> not_loaded <- FALSE
#> }
#> } 
#>  
#>  rand_art <- remDr$findElement(using = 'id', 'n-randompage')
#> rand_art$clickElement()
#> Sys.sleep(0.25) 
#>  
#>  title <- try(remDr$findElement(using = 'id', 'firstHeading')) 
#> if(is(title,'try-error')){ 
#> title <- NA 
#> } else { 
#> title <- title$getElementText()[[1]] 
#> } 
#>  
#>  link <- remDr$findElement(using = 'xpath', '/html/body/div[3]/div[3]/div[5]/div[1]/p[1]/a[1]')
#> link$clickElement()
#> Sys.sleep(0.25) 
#>  
#>  first_link_title <- try(remDr$findElement(using = 'id', 'firstHeading')) 
#> if(is(first_link_title,'try-error')){ 
#> first_link_title <- NA 
#> } else { 
#> first_link_title <- first_link_title$getElementText()[[1]] 
#> } 
#>  
#>  first_link_text <- try(remDr$findElement(using = 'xpath', '/html/body/div[3]/div[3]/div[5]/div[1]/p[1]')) 
#> if(is(first_link_text,'try-error')){ 
#> first_link_text <- NA 
#> } else { 
#> first_link_text <- first_link_text$getElementText()[[1]] 
#> }