Skip to content

ropensci/rtika

Repository files navigation

rtika

Extract text or metadata from over a thousand file types.

R-CMD-check ROpenSci Coverage status Cranlogs Downloads

Apache Tika is a content detection and analysis framework, written in Java, stewarded at the Apache Software Foundation. It detects and extracts metadata and text from over a thousand different file types, and as well as providing a Java library, has server and command-line editions suitable for use from other programming languages …

For most of the more common and popular formats, Tika then provides content extraction, metadata extraction and language identification capabilities. (From https://en.wikipedia.org/wiki/Apache_Tika, accessed Jan 18, 2018)

This is an R interface to the Tika software.

Installation

To start, you need R and Java 8 or OpenJDK 1.8. Higher versions work. To check your version, run the command java -version from a terminal. Get Java installation tips at https://www.java.com/en/download/ or https://openjdk.org/install/. Because the rJava package is not required, installation is simple. You can cut and paste the following snippet:

install.packages('rtika', repos = 'https://cloud.r-project.org')

library('rtika')

# You need to install the Apache Tika .jar once.
install_tika()

Read an introductory article at https://docs.ropensci.org/rtika/articles/rtika_introduction.html.

Key Features

  • tika_text() to extract plain text.
  • tika_xml() and tika_html() to get a structured XHMTL rendition.
  • tika_json() to get metadata as .json, with XHMTL content.
  • tika_json_text() to get metadata as .json, with plain text content.
  • tika() is the main function the others above inherit from.
  • tika_fetch() to download files with a file extension matching the Content-Type.

Supported File Types

Tika parses and extracts text or metadata from over one thousand digital formats, including:

  • Portable Document Format (.pdf)
  • Microsoft Office document formats (Word, PowerPoint, Excel, etc.)
  • Rich Text Format (.rtf)
  • Electronic Publication Format (.epub)
  • Image formats (.jpeg, .png, etc.)
  • Mail formats (.mbox, Outlook)
  • HyperText Markup Language (.html)
  • XML and derived formats (.xml, etc.)
  • Compression and packaging formats (.gzip, .rar, etc.)
  • OpenDocument Format
  • iWorks document formats
  • WordPerfect document formats
  • Text formats
  • Feed and Syndication formats
  • Help formats
  • Audio formats
  • Video formats
  • Java class files and archives
  • Source code
  • CAD formats
  • Font formats
  • Scientific formats
  • Executable programs and libraries
  • Crypto formats

For a list of MIME types, look for the “Supported Formats” page here: https://tika.apache.org/

Get Plain Text

The rtika package processes batches of documents efficiently, so I recommend batches. Currently, the tika() parsers take a tiny bit of time to spin up, and that will get annoying with hundreds of separate calls to the functions.

# Test files
batch <- c(
  system.file("extdata", "jsonlite.pdf", package = "rtika"),
  system.file("extdata", "curl.pdf", package = "rtika"),
  system.file("extdata", "table.docx", package = "rtika"),
  system.file("extdata", "xml2.pdf", package = "rtika"),
  system.file("extdata", "R-FAQ.html", package = "rtika"),
  system.file("extdata", "calculator.jpg", package = "rtika"),
  system.file("extdata", "tika.apache.org.zip", package = "rtika")
)

# batches are best, and can also be piped with magrittr.
text <- tika_text(batch)

# text has one string for each document:
length(text)
#> [1] 7

# A snippet:
cat(substr(text[1], 54, 190)) 
#> lite’
#> June 1, 2017
#> 
#> Version 1.5
#> 
#> Title A Robust, High Performance JSON Parser and Generator for R
#> 
#> License MIT + file LICENSE
#> 
#> NeedsCompi

To learn more and find out how to extract structured text and metadata, read the vignette: https://docs.ropensci.org/rtika/articles/rtika_introduction.html.

Enhancements

Tika also can interact with the Tesseract OCR program on some Linux variants, to extract plain text from images of text. If tesseract-ocr is installed, Tika should automatically locate and use it for images and PDFs that contain images of text. However, this does not seem to work on OS X or Windows. To try on Linux, first follow the Tesseract installation instructions. The next time Tika is run, it should work. For a different approach, I suggest tesseract package by @jeroen, which is a specialized R interface.

The Apache Tika community welcomes your feedback. Issues regarding the R interface should be raised at the rTika Github Issue Tracker. If you are confident the issue concerns Tika or one of its underlying parsers, use the Tika Bugtracking System.

Using the Tika App Directly

If your project or package needs to use the Tika App .jar, you can include rTika as a dependency and call the rtika::tika_jar() function to get the path to the Tika app installed on the system.

Similar R Packages

The are a number of specialized parsers that overlap in functionality. For example, the pdftools package extracts metadata and text from PDF files, the antiword package extracts text from recent versions of Word, and the epubr package by @leonawicz processes epub files. These packages do not depend on Java, while rTika does.

The big difference between Tika and a specialized parser is that Tika integrates dozens of specialist libraries maintained by the Apache Foundation. Apache Tika processes over a thousand file types and multiple versions of each. This eases the processing of digital archives that contain unpredictable files. For example, researchers use Tika to process archives from court cases, governments, or the Internet Archive that span multiple years. These archives frequently contain diverse formats and multiple versions of each format. Because Tika finds the matching parser for each individual file, is well suited to diverse sets of documents. In general, the parsing quality is good and consistently so. In contrast, specialized parsers may only work with a particular version of a file, or require extra tinkering.

On the other hand, a specialized library can offer more control and features when it comes to structured data and formatting. For example, the tabulizer package by @leeper and @tpaskhalis includes bindings to the ‘Tabula PDF Table Extractor Library’. Because PDF files store tables as a series of positions with no obvious boundaries between data cells, extracting a data.frame or matrix requires heuristics and customization which that package provides. To be fair to Tika, there are some formats where rtika will extract data as table-like XML. For example, with Word and Excel documents, Tika extracts simple tables as XHTML data that can be turned into a tabular data.frame using the rvest::html_table() function.

History

In September 2017, github.com user kyusque released tikaR, which uses the rJava package to interact with Tika (See: https://github.com/kyusque/tikaR). As of writing, it provided similar text and metadata extraction, but only xml output.

Back in March 2012, I started a similar project to interface with Apache Tika. My code also used low-level functions from the rJava package. I halted development after discovering that the Tika command line interface (CLI) was easier to use. My empty repository is at https://r-forge.r-project.org/projects/r-tika/.

I chose to finally develop this package after getting excited by Tika’s new ‘batch processor’ module, written in Java. The batch processor has very good efficiency when processing tens of thousands of documents. Further, it is not too slow for a single document either, and handles errors gracefully. Connecting R to the Tika batch processor turned out to be relatively simple, because the R code is simple. It uses the CLI to point Tika to the files. Simplicity, along with continuous testing, should ease integration. I anticipate that some researchers will need plain text output, while others will want json output. Some will want multiple processing threads to speed things up. These features are now implemented in rtika, although apparently not in tikaR yet.

Code of Conduct

Please note that this project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms.

ropensci_footer