R-binding

MITIE R Package

MITIE's functionality can be used in the R statistical computing environment.

Linux users can download (or build) the MITIE source package, and then install the package into R from source.

Windows users should download the pre-built binary package and install it through an R graphical user interface.

Building the Source Package

The MITIE source package is available for download. However, it can be packaged with the following steps.

Starting in the top level MITIE folder, run the commands:

cd tools/R-binding
./copy_source.sh
./build_source_package.sh

This will create a MITIE_{version}.tar.gz source package.

Installing the MITIE Source Package (Linux/Mac)

Note: The following requires the R development tools to be installed, e.g., the r-base-dev package on ubuntu.

Step 1: Install MITIE dependencies.

MITIE depends on the Rcpp package for integrating C++ with R. Rcpp can be installed using a GUI package manager or using the R command:

install.packages("Rcpp")

Step 2: Build or download latest MITIE source package.

Step 3: From the linux command line, run:

R CMD INSTALL MITIE_{version}.tar.gz

Installing the MITIE Binary Package (Windows)

The MITIE package can be installed from source on Windows if Rtools is installed. However, pre-compiled Windows binaries are also available (and more convenient).

Step 1: Install MITIE dependencies.

MITIE depends on the Rcpp package for integrating C++ with R. Rcpp can be installed using a GUI package manager or using the R command:

install.packages("Rcpp")

Step 2: Download the latest Windows binary package, which was built with R version 3.1.

Step 3: Use GUI's package manager to install package from local zip file. For example, in RGui, select "Packages" / "Install package(s) from local zip files..." and navigate to MITIE_{version}.zip to install the package.

Using MITIE from R

MITIE requires trained model files to do named entity extraction, binary relation extraction, etc. Models for English and Spanish are currently available: MITIE-models-v0.2.tar.bz2 and MITIE-models-v0.2-Spanish.zip.

The following R session demonstrates how to perform named entity extraction.

library(MITIE)
help(MITIE)

# Load named entity extractor from disk
# NOTE: models can be downloaded from http://sourceforge.net/projects/mitie/files/binaries/
# NOTE: change this path to point to where your model files are

ner_model_path <- "C:/MITIE-models/english/ner_model.dat"
ner <- NamedEntityExtractor$new(ner_model_path)

# Print out what kind of tags this tagger can predict

tag_names <- ner$get_possible_ner_tags()
tag_names

#  [1] "PERSON"       "LOCATION"     "ORGANIZATION" "MISC"

# Prepare some data

tokens <- mitie_tokenize("Bill Gates was born in Seattle, Washington. Bill used to be the CEO of Microsoft.")
tokens

#  [1] "Bill"       "Gates"      "was"        "born"       "in"        
#  [6] "Seattle"    ","          "Washington" "."          "Bill"      
# [11] "used"       "to"         "be"         "the"        "CEO"       
# [16] "of"         "Microsoft"  "."  

# Extract entities

entities <- ner$extract_entities(tokens)
for (i in 1:length(entities)) {
    entity = entities[[i]]
    position = paste("(", entity$start, ",", entity$end, ")", sep="")
    text = paste(tokens[entity$start:entity$end], collapse=" ")
    print(paste(text, "/", tag_names[entity$tag], "@", position))
}

# [1] "Bill Gates / PERSON @ (1,2)"
# [1] "Seattle / LOCATION @ (6,6)"
# [1] "Washington / LOCATION @ (8,8)"
# [1] "Bill / PERSON @ (10,10)"
# [1] "Microsoft / ORGANIZATION @ (17,17)"