Crawling through a web labyrinth using {rvest}

The data, the website

The data source contains a school register. The data is located behind this antique-looking form. The bit we want – for a Covid-19-related project – is contact information for each of the country’s several thousand primary and secondary education institutions, as well as some metadata about the type of school, who established it etc.

The path to the data

The thing we want is not to scrape the massive paginated table we get from the search form, but to download the “Excel” (more on that later) export you can get after you run the search query. Unfortunately that is hidden behind two layers of forms and is itself downloaded by submitting one of four forms in the last layer.

The whole thing is written in some ASP/.Net tools I do not understand – all I can tell from inspecting the requests in browser developer tools and from trying to replicate those requests in curl, httr and in simple rvest steps is that the whole thing is held together by a bunch of cookies and session logics which the client needs to keep track of and return exactly, otherwise the server throws errors.

The rvest workflow

{rvest} to the rescue. One set of tools inside {rvest} that I previously ignored is the session-based workflow – which happens to solve exactly this problem. I started playing with it just as Hadley was doing some improvements and refactoring, so here is my understanding of how it works currently (it is now in the dev version on Github but probably soon on CRAN):

library(rvest)

1. Start a session using session():

sess <- session("http://stistko.uiv.cz/registr/vybskolrn.asp")
sess
## <session> http://stistko.uiv.cz/registr/vybskolrn.asp
##   Status: 200
##   Type:   text/html
##   Size:   23117

This is an rvest_session object but can also be approached as a html_document. E.g. you can extract a form from it:

2. Extract the form

form1 <- html_form(sess)

The benefit of having the session object is that you can use it to submit the form, and all the right cookies will be automatically submitted with it.

Note that we added a user agent header to identify ourselves.

3. Fill the form

But first, we need to set the values in the form – as it turns out, the defaults for this particular form, as seen in the browser, are not actually encoded in the HTML, so even if we want to submit a blank “give me everything” search, we need to manually set the defaults. (The object returned by html_form() is a list of rvest_form objects, which we nead to bear in mind when inspecting it.)

e.g. let’s look at a particular field:

form1[[1]][['fields']][['uzemi']]
## <field> (select) uzemi:

and see its value (normally that would be the default):

form1[[1]][['fields']][['uzemi']]$value
## character(0)

So to set the values in the form, we need to have a look at what the first option is and set that as the value.

form1[[1]][['fields']][['uzemi']][['options']][[1]]
## [1] "NIC"

That turns out to be the case for most of the select fields, so we can set it like so:

form1set <- html_form_set(form1[[1]], 
                          # set to a small region so we don't blast the 
                          # server with heavy queries
                          uzemi = "CZ0514", 
                          # the rest are defaults ("NOTHING")
                          zrizovatel = "NIC",
                          organ = "NIC",
                          typ = "NIC",
                          jazs = "NIC",
                          delka = "NIC",
                          forma = "NIC",
                          jazob = "NIC",
                          skupobor = "NIC",
                          kmobor = "NIC",
                          obor = "NIC")

4. Submit the form, using the session

session_after_form1 <- session_submit(sess, form1set, submit = NULL, httr::user_agent("github.com/petrbouchal"))
## Submitting with 'XX'
session_after_form1
## <session> http://stistko.uiv.cz/registr/skolyrn.asp
##   Status: 200
##   Type:   text/html
##   Size:   33316

Now if we look at the results page in browser dev tools, we will see that the way to proceed to the export page is to submit another form:

results_forms <- session_after_form1 %>% html_form()
length(results_forms)
## [1] 14
results_forms[1]
## [[1]]
## <form> '<unnamed>' (POST http://stistko.uiv.cz/registr/vybskolrn.asp)
##   <field> (submit) AA: Návrat na výběr
results_forms[2]
## [[1]]
## <form> '<unnamed>' (POST http://stistko.uiv.cz/registr/exportqx.asp)
##   <field> (submit) EX: Export do Excelu

Fourteen forms on a simple results page…

The text suggests the second form is what we want – it’s a single button, so we submit it. Notice that (a) we use session_submit() and (b) an object from the previous call is again included in the submit call to maintain the session:

export_page <- session_submit(session_after_form1, results_forms[[2]], submit = NULL,
                              httr::user_agent("github.com/petrbouchal"))
export_page
## <session> http://stistko.uiv.cz/registr/exportqx.asp
##   Status: 200
##   Type:   text/html
##   Size:   2782

Looking at this in browser, we find we now face another set of forms which will hopefully give us the coveted excel exports.

export_forms <- html_form(export_page)
export_forms
## [[1]]
## <form> '<unnamed>' (POST http://stistko.uiv.cz/registr/exportq.asp)
##   <field> (submit) EX: Export adresáře
## 
## [[2]]
## <form> '<unnamed>' (POST http://stistko.uiv.cz/registr/exportqs.asp)
##   <field> (submit) EXX: Export škol
## 
## [[3]]
## <form> '<unnamed>' (POST http://stistko.uiv.cz/registr/exportqm.asp)
##   <field> (submit) EXM: Export škol s mís...
## 
## [[4]]
## <form> '<unnamed>' (POST http://stistko.uiv.cz/registr/exportqo.asp)
##   <field> (submit) EXO: Export škol s obo...

These are exports of different bits of data from the database. Let’s look at the first one:

export <- session_submit(export_page, export_forms[[1]], submit = NULL,
                         httr::user_agent("github.com/petrbouchal"))
export
## <session> http://stistko.uiv.cz/registr/exportq.asp
##   Status: 200
##   Type:   application/vnd.ms-excel
##   Size:   55674

The quasi excel data dump

This seems to be an excel file of 55 kB – great. What exactly is it?

export$response$headers$`content-disposition`
## [1] "attachment; filename=Adresar.xls"

An XLS file to be precise. Let’s write it to disk.

writeBin(export$response$content, "export.xls")

And read it in:

readxl::read_excel("export.xls")
## Error: 
##   filepath: /Users/petr/cpers/petrbouchal.github.io/content/post/2021-01-26-rvesting/export.xls
##   libxls error: Unable to open file

Hmm…what it this?

readLines("export.xls", n = 3)
## [1] ""       "<html>" "<head>"

Right, so this is actually a HTML file. Is it a table?

stringr::str_sub(readr::read_lines("export.xls", n_max = 1, skip = 7), 1, 100)
## [1] "<table border='1'><tr><td>RED_IZO</td><td>I\xc8O</td><td>Z\xf8izovatel</td><td>\xdazem\xed</td><td>Kraj</td><td>"

Let’s try parsing it then.

tbl <- read_html("export.xls") %>% 
  html_table(header = T)
tbl[[1]]
## # A tibble: 99 x 27
##    RED_IZO    IČO Zřizovatel Území Kraj  `Správní úřad`   ORP `Plný název`
##      <int>  <int>      <int> <chr> <chr> <chr>          <int> <chr>       
##  1  6.00e8 8.55e5          7 CZ05… Libe… B51000          5109 Gymnázium, …
##  2  6.00e8 8.55e5          7 CZ05… Libe… B51000          5109 Střední umě…
##  3  6.00e8 8.56e5          7 CZ05… Libe… B51000          5107 Gymnázium I…
##  4  6.00e8 5.81e5          7 CZ05… Libe… B51000          5109 Střední zdr…
##  5  6.00e8 7.08e7          7 CZ05… Libe… B51000          5107 Základní šk…
##  6  6.00e8 8.55e5          7 CZ05… Libe… B51000          5107 Dětský domo…
##  7  6.00e8 7.09e7          7 CZ05… Libe… B51000          5107 Pedagogicko…
##  8  6.00e8 7.10e7          2 CZ05… Libe… D51040          5104 Mateřská šk…
##  9  6.00e8 7.07e7          2 CZ05… Libe… D51080          5108 Mateřská šk…
## 10  6.00e8 7.27e7          2 CZ05… Libe… D51040          5104 Mateřská šk…
## # … with 89 more rows, and 19 more variables: `Zkrácený název` <chr>,
## #   Ulice <chr>, č.p. <int>, č.or. <lgl>, č.obce <chr>, PSČ <int>, Místo <chr>,
## #   Telefon <chr>, Fax <chr>, `Email 1` <chr>, `Email 2` <chr>, WWW <chr>, `ID
## #   dat. schránky subjektu` <chr>, Ředitel <chr>, X <chr>, `je OVM` <chr>,
## #   ZUJ <int>, `Email zřizovatele` <chr>, `ID dat. schránky zřizovatele` <chr>

There we go – so the {rvest} machinery turned out to be useful not just for getting through the maze of forms and cookies, but for parsing a HTML table masquerading as an Excel file.

If you are interested in how this is packaged into a (very early stages) package, take a look at {vsezved}: code, website.