knitr::opts_chunk$set( collapse = TRUE, comment = "#>", message = FALSE, warning = FALSE, error = FALSE, tidy = TRUE ) library(dplyr)
epubr package provides functions supporting the reading and parsing of internal e-book content from EPUB files. E-book metadata and text content are parsed separately and joined together in a tidy, nested tibble data frame.
E-book formatting is not completely standardized across all literature. It can be challenging to curate parsed e-book content across an arbitrary collection of e-books perfectly and in completely general form, to yield a singular, consistently formatted output. Many EPUB files do not even contain all the same pieces of information in their respective metadata.
EPUB file parsing functionality in this package is intended for relatively general application to arbitrary EPUB e-books. However, poorly formatted e-books or e-books with highly uncommon formatting may not work with this package.
There may even be cases where an EPUB file has DRM or some other property that makes it impossible to read with
Text is read 'as is' for the most part. The only nominal changes are minor substitutions, for example curly quotes changed to straight quotes. Substantive changes are expected to be performed subsequently by the user as part of their text analysis. Additional text cleaning can be performed at the user's discretion, such as with functions from packages like
Bram Stoker's Dracula novel sourced from Project Gutenberg is a good example of an EPUB file with unfortunate formatting.
The first thing that stands out is the naming convention using
item followed by some ordered digits does not differentiate sections like the book preamble from the chapters.
The numbering also starts in a weird place. But it is actually worse than this. Notice that sections are not broken into chapters; they can begin and end in the middle of chapters!
These annoyances aside, the metadata and contents can still be read into a convenient table. Text mining analyses can still be performed on the overall book, if not so easily on individual chapters. See the section below on restructuring for examples of
epubr functions that help get around these issues.
Here a single file is read with
epub(). The output of the returned primary data frame and the book text data frame that is nested within its
data column are shown.
library(epubr) file <- system.file("dracula.epub", package = "epubr") (x <- epub(file)) x$data[]
file argument may be a vector of EPUB files. There is one row for each book.
The above examples jump right in, but it can be helpful to inspect file metadata before reading a large number of books into memory. Formatting may differ across books. It can be helpful to know what fields to expect, the degree of consistency, and what content you may want to drop during the file reading process.
epub_meta() strictly parses file metadata and does not read the e-book text.
This provides the big picture, though it will not reveal the internal breakdown of book section naming conventions that were seen in the first
file can also be a vector for
file is a vector, the fields (columns) returned are the union of all fields detected across all EPUB files. Any books (rows) that do not have a field found in another book return
NA for that row and column.
There are three optional arguments that can be provided to
Unless you have a collection of well-formatted and similarly formatted EPUB files, these arguments may not be helpful and can be ignored, especially chapter detection.
Selecting fields is straightforward. All fields found are returned unless a vector of fields is provided.
epub(file, fields = c("title", "creator", "file"))
file was not a field identified in the metadata. This is a special case. Including
file will include the
basename of the input file. This is helpful when you want to retain file names and
source is included in the metadata but may represent something else. Some fields like
title are always returned and do not need to be specified in
Also, if your e-book does not have a metadata field named
title, you can pass an additional argument to
... to map a different, known metadata field to
title = "BookTitle". The resulting table always has a
title field, but in this case
title would be populated with information from the
BookTitle metadata field. If the default
title field or any other field name passed to the additional
title argument does not exist in the file metadata, the output
title column falls back on filling in with the same unique file names obtained when requesting the
Filtering out unwanted sections, or rows of the nested data frame, uses a regular expression pattern. Matched rows are dropped. This is where knowing the naming conventions used in the e-books in
file, or at least knowing they are satisfactorily consistent and predictable for a collection, helps with removing extraneous clutter.
One section that can be discarded is the cover. For many books it can be helpful to use a pattern like
"^(C|c)ov" to drop any sections whose IDs begin with
cov, and may be that abbreviation or the full word. For this book,
cov suffices. The nested data frame has one less row than before.
epub(file, drop_sections = "cov")$data[]
This e-book unfortunately does not have great formatting. For the sake of example, pretend that chapters are known to be sections beginning with
item and followed by two digits, using the pattern
^item\\d\\d. This does two things. It adds a new metadata column to the primary data frame called
nchap giving the estimated number of chapters in the book. In the nested data frame containing the parsed e-book text, the
section column is conditionally mutated to reflect a new, consistent chapter naming convention for the identified chapters and a logical
is_chapter column is added.
x <- epub(file, drop_sections = "cov", chapter_pattern = "^item\\d\\d") x x$data[]
This renaming of sections is helpful if the e-book content is split into meaningful sections, but the sections are not named in a similarly meaningful way. It is not helpful in cases where the sections are poorly or arbitrarily defined as in the above example where they begin with
item* regardless of content. See the examples below on restructuring parsed content for an approach to dealing with this more problematic case.
Another problematic formatting that you may encounter is the random order of book sections as defined in the metadata, even if the sections themselves are well-defined. For example, there may be clear breaks between chapters, but the chapters may occur out of order such as
ch05, ch11, ch02, ... and so on. Merely identifying which sections are chapters does not help in this case. They are still be labeled incorrectly since they are already out of order. Properly reordering sections is also addressed below.
Also note that not all books have chapters or something like them. Make sure an optional argument like
chapter_pattern makes sense to use with a given e-book in the first place.
Ultimately, everything depends on the quality of the EPUB file. Some publishers are better than others. Formatting standards may also change over time.
When reading EPUB files it is ideal to be able to identify meaningful sections to retain via a regular expression pattern, as well as to drop extraneous sections in a similar manner. Using pattern matching as shown above is a convenient way to filter rows of the nested text content data frame.
For e-books with poor metadata formatting this is not always possible, or may be possible only after some other pre-processing.
epubr provides other functions to assist in restructuring the text table. The Dracula EPUB file included in
epubr is a good example to continue with here.
This book is internally broken into sections at arbitrary break points, hence why several sections begin in the middle of chapters, as seen above. Other chapters begin in the middle of sections. Use
epub_recombine() along with a regular expression that can match the true section breaks. This function collapses the full text and then rebuilds the text table using new sections with proper break points. In the process it also recalculates the numbers of words and characters and relabels the sections with chapter notation.
Fortunately, a reliable pattern exists, which consists of
CHAPTER in capital letters followed by a space and some Roman numerals. Recombine the text into a new object.
pat <- "CHAPTER [IVX]+" x2 <- epub_recombine(x, pat) x2 x2$data[]
But this is not quite as expected. There should be 27 chapters, not 54. What was not initially apparent was that the same pattern matching each chapter name also appears in the first section where every chapter is listed in the table of contents. The new section breaks were successful in keeping chapter text in single, unique sections, but there are now twice as many as needed. Unintentionally, the first 27 "chapters" represent the table of contents being split on each chapter ID. These should be removed.
An easy way to do this is with
epub_sift(), which sifts, or filters out, small word- or character-count sections from the nested data frame. It's a simple sieve and you can control the size of the holes with
n. You can choose
type = "word" (default) or
type = "character". This is somewhat of a blunt instrument, but is useful in a circumstance like this one where it is clear it will work as desired.
library(dplyr) x2 <- epub_recombine(x, pat) %>% epub_sift(n = 200) x2 x2$data[]
This removes the unwanted rows, but one problem remains. Note that sifting the table sections in this case results in a need to re-apply
epub_recombine() because the sections we removed had nevertheless offset the chapter indexing. Another call to
epub_recombine() can be chained, but it may be more convenient to use the
sift argument to
epub_recombine(), which is applied recursively.
#epub_recombine(x, pat) %>% epub_sift(n = 200) %>% epub_recombine(pat) x2 <- epub_recombine(x, pat, sift = list(n = 200)) x2 x2$data[]
Some poorly formatted e-books have their internal sections occur in an arbitrary order. This can be frustrating to work with when doing text analysis on each section and where order matters. Just like recombining into new sections based on a pattern, sections that are out of order can be reordered based on a pattern. This requires a bit more work. In this case the user must provide a function that will map something in the matched pattern to an integer representing the desired row index.
Continue with the Dracula example, but with one difference. Even though the sections were originally broken at arbitrary points, they were in chronological order. To demonstrate the utility of
epub_reorder(), first randomize the rows so that chronological order can be recovered.
set.seed(1) x2$data[] <- sample_frac(x2$data[]) # randomize rows for example x2$data[]
It is clear above that sections are now out of order. It is common enough to load poorly formatted EPUB files and yield this type of result. If all you care about is the text in its entirely, this does not matter, but if your analysis involves trends over the course of a book, this is problematic.
For this book, you need a function that will convert an annoying Roman numeral to an integer. You already have the pattern for finding the relevant information in each text section. You only need to tweak it for proper substitution. Here is an example:
f <- function(x, pattern) as.numeric(as.roman(gsub(pattern, "\\1", x)))
This function is passed to
epub_reorder(). It takes and returns scalars. It must take two arguments: the first is a text string. The second is the regular expression. It must return a single number representing the index of that row. For example, if the pattern matches
CHAPTER IV, the function should return a
epub_reorder() takes care of the rest. It applies your function to every row in the the nested data frame and then reorders the rows based on the full set of indices. Note that it also repeats this for every row (book) in the primary data frame, i.e., for every nested table. This means that the same function will be applied to every book. Therefore, you should only use this in bulk on a collection of e-books if you know the pattern does not change and the function will work correctly in each case.
The pattern has changed slightly. Parentheses are used to retain the important part of the matched pattern, the Roman numeral. The function
f here substitutes the entire string (because now it begins with
^ and ends with
.*) with only the part stored in parentheses (In
f, this is the
epub_reorder() applies this to all rows in the nested data frame:
x2 <- epub_reorder(x2, f, "^CHAPTER ([IVX]+).*") x2$data[]
It is important that this is done on a nested data frame that has already been cleaned to the point of not containing extraneous rows that cannot be matched by the desired pattern. If they cannot be matched, then it is unknown where those rows should be placed relative to the others.
If sections are both out of order and use arbitrary break points, it would be necessary to reorder them before you split and recombine. If you split and recombine first, this would yield new sections that contain text from different parts of the e-book. However, the two are not likely to occur together; in fact it may be impossible for an EPUB file to be structured this way. In developing
epubr, no such examples have been encountered. In any event, reordering out of order sections essentially requires a human-identifiable pattern near the beginning of each section text string, so it does not make sense to perform this operation unless the sections have meaningful break points.
Separate from using
epub(), you can call
epub_unzip() directly if all you want to do is extract the files from the
.epub file archive. By default the archive files are extracted to
tempdir() so you may want to change this with the
bookdir <- file.path(tempdir(), "dracula") epub_unzip(file, exdir = bookdir) list.files(bookdir, recursive = TRUE)
The helper function
count_words() provides word counts for strings, but allows you to control the regular expression patterns used for both splitting the string and conditionally counting the resulting character elements. This is the same function used internally by
epub_recombine(). It is exported so that it can be used directly.
count_words() splits on spaces and new line characters. It counts as a word any element containing at least one alphanumeric character or the ampersand. It ignores everything else as noise, such as extra spaces, empty strings and isolated bits of punctuation.
x <- " This sentence will be counted to have:\n\n10 (ten) words." count_words(x)
Helper functions for inspecting the text in the R console include
epub_head() provides an overview of the text by section for each book in the primary data frame. The nested data frames are unnested and row bound to one another and returned as a single data frame. The text is shortened to only the first few characters (defaults to
n = 50).
epub_cat() can be used to
cat the text of an e-book to the console for quick inspection in a more readable form. It can take several arguments that help slice out a section of the text and customize how it is printed.
Both functions can take an EPUB filename or a data frame of an already loaded EPUB file as their first argument.
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.