PDF scraping: get company and subsidiaries tables

I am trying to scrape this PDF containing information about company subsidiaries. I have seen many posts using the R package Tabulizer but this, unfortunately, doesn't work on my Mac for some reasons. As Tabulizer uses Java dependencies, I tried installing different versions of Java (6-13) and then reinstalling the packages, still no luck in getting this to work (what happens is when I run extract_tables the R session aborts).

I need to scrape the whole pdf from page 19 onwards and construct a table showing company names and their subsidiaries. In the pdf, names start with any letters/number/symbol, whereas subsidiaries start with either a single or double dot.

So I tried with pdftools and pdftables packages. The code below provides a table similar to the one on page 19:

library(pdftools)
library(pdftables)
library(tidyverse)

tt = pdf_text("~/DATA/978-1-912036-41-7-Who Owns Whom UK-Ireland-Volume-1.pdf")

df <- tt[19]
df2 <- strsplit(df, '  ')
 
df3 <-as.data.frame(do.call(cbind, df2)) %>% 
 filter(V1!="") %>% 
 mutate(V2=str_split_fixed(V1, "England . ", 2)) %>% 
 mutate(V3=str_split_fixed(V1, "England", 2)) %>% 
 select(V2,V3,V1) %>% 
 mutate(V1=ifelse(V1==V3,"",V1),V3=ifelse(V3==V2,"",V3)) %>% 
 select(V3,V2,V1) %>% 
 mutate_at(c("V1"), funs(lead), n = 1 ) %>% 
 mutate_at(c("V3"), funs(lag), n = 1 ) %>% 
 unite(V4,V1, V2, V3, sep = "", remove = FALSE)

I am sure there is a more sophisticated function to do this more neatly. For example by using '\n' or '\r' with strsplit:

 df2 <- strsplit(df, '\n') 
 df3 <- do.call(cbind.data.frame, df2)

Can anyone with more experience than me advise me on how to scrape this table?

Solution

Like @Justin Coco hinted, this was a lot of fun. The code ended up a bit more complex than I anticipated, but I think the result should be what you imagined.

I used pdf_data instead of pdf_text so I can work with the position of words.

library(pdftools)
#> Using poppler version 0.86.1
library(tidyverse)
pdf_location <- "/location/of/pdf"
pdf_raw <- pdf_data(pdf_location)

I then wrote a function which can process a page from the PDF:

get_table <- function(x, page) {
  x[[page]] %>% # select page, I use this variable again below, which is why I'm not simply looping through the whole object

    filter(y > 25, y < 833) %>% # above and below these positions is the pdf header which we are not interested in
    mutate(column = case_when( # I check the x-positions where the columns start an end and transformed them into column numbers
      x >= 36 & x < 220 ~ 1L,
      x >= 220 & x < 403 ~ 2L,
      x >= 403 ~ 3L,
    )) %>% 
    mutate(newrow = case_when( # check if this is a new line
      column == 1L & x == 36  ~ TRUE, 
      column == 2L & x == 220 ~ TRUE,
      column == 3L & x == 403 ~ TRUE,
      TRUE ~ FALSE
    ),
    row = cumsum(newrow), # get the row number
    subsidiary = newrow & text == ".") %>% # as you say, subsidiary names start with "."
    group_by(row, column) %>% # grouping and summarising moves the text into one 'cell'
    summarise(text = paste(text, collapse = " "), 
              subsidiary = sum(subsidiary) > 0,
              .groups = "drop") %>% 
    mutate(headline = !str_detect(text, "\\s")) %>% # the category headlines (@, A, B, C, etc.) are still in there but can be identified easily since they lack whitespace
    mutate(row = ifelse(row > 1 & !subsidiary & !lag(subsidiary) & !lag(headline), lag(row), row),
           row = ifelse(row > 1 & !subsidiary & !lag(subsidiary) & !lag(headline), lag(row), row)) %>% # some company names stretch over up to three lines but lines are not indented
    group_by(row, column) %>% 
    summarise(text = paste(text, collapse = " "), 
              subsidiary = sum(subsidiary) > 0,
              headline = head(headline, 1),
              .groups = "drop")  %>% 
    
    mutate(page = page, .before = row) # finally add the page number to keep track
}

You can test this on one page or loop through all of them at once:

pdf_df <- map_df(19:1428, ~get_table(pdf_raw, page = .x))

I already like the df, but you requested that the table should be "showing company names and their subsidiaries". So let's do some more wrangling on the pdf_df object.

pdf_df %>% 
  filter(!headline) %>% 
  mutate(company_nr = cumsum(!subsidiary)) %>% 
  group_by(company_nr) %>% 
  mutate(company = text[!subsidiary & !headline]) %>% 
  filter(subsidiary) %>% 
  select(company_nr, company, subsidiary = text)
#> # A tibble: 303,380 x 3
#> # Groups:   company_nr [115,477]
#>    company_nr company                             subsidiary                    
#>         <int> <chr>                               <chr>                         
#>  1          1 ?WHAT IF! HOLDINGS LIMITED The Gla… . ?What If! China Holdings Li…
#>  2          1 ?WHAT IF! HOLDINGS LIMITED The Gla… . . ?What If! Innovation Sing…
#>  3          1 ?WHAT IF! HOLDINGS LIMITED The Gla… . ?What If! Joint Ventures Li…
#>  4          1 ?WHAT IF! HOLDINGS LIMITED The Gla… . ?What If! Limited England   
#>  5          1 ?WHAT IF! HOLDINGS LIMITED The Gla… . . ? What If ! Inventors Lim…
#>  6          1 ?WHAT IF! HOLDINGS LIMITED The Gla… . . ? What If ! Training Limi…
#>  7          1 ?WHAT IF! HOLDINGS LIMITED The Gla… . Nobby Styles Limited Englan…
#>  8          2 @A COMPANY LIMITED Premier Suite 4… . Aviva Holdings Limited Engl…
#>  9          2 @A COMPANY LIMITED Premier Suite 4… . Copper Mountain Networks Li…
#> 10          2 @A COMPANY LIMITED Premier Suite 4… . Just Ties Limited England   
#> # … with 303,370 more rows

^{Created on 2021-05-23 by the reprex package (v2.0.0)}

Let me know in a comment if there are problems. I obviously didn't go through all pages to check if the script has some quirks with specific company names etc. but the first pages look fine to me.