MITIE: MIT Information Extraction
This project provides free (even for commercial use) state-of-the-art information extraction tools. The current release includes tools for performing named entity extraction and binary relation detection as well as tools for training custom extractors and relation detectors.
MITIE is built on top of dlib, a high-performance machine-learning library[1], MITIE makes use of several state-of-the-art techniques including the use of distributional word embeddings[2] and Structural Support Vector Machines[3]. MITIE offers several pre-trained models providing varying levels of support for both English, Spanish, and German trained using a variety of linguistic resources (e.g., CoNLL 2003, ACE, Wikipedia, Freebase, and Gigaword). The core MITIE software is written in C++, but bindings for several other software languages including Python, R, Java, C, and MATLAB allow a user to quickly integrate MITIE into his/her own applications.
Outside projects have created API bindings for OCaml, .NET, .NET Core, and Ruby. There is also an interactive tool for labeling data and training MITIE.
Using MITIE
MITIE's primary API is a C API which is documented in the mitie.h header file. Beyond this, there are many example programs showing how to use MITIE from C, C++, Java, R, or Python 2.7.
Initial Setup
Before you can run the provided examples you will need to download the trained model files which you can do by running:
make MITIE-models
or by simply downloading the MITIE-models-v0.2.tar.bz2 file and extracting it in your MITIE folder. Note that the Spanish and German models are supplied in separate downloads. So if you want to use the Spanish NER model then download MITIE-models-v0.2-Spanish.zip and extract it into your MITIE folder. Similarly for the German model: MITIE-models-v0.2-German.tar.bz2
Using MITIE from the command line
MITIE comes with a basic streaming NER tool. So you can tell MITIE to process each line of a text file independently and output marked up text with the command:
cat sample_text.txt | ./ner_stream MITIE-models/english/ner_model.dat
The ner_stream executable can be compiled by running make in the top level MITIE folder or by navigating to the tools/ner_stream folder and running make or using CMake to build it which can be done with the following commands:
cd tools/ner_stream
mkdir build
cd build
cmake ..
cmake --build . --config Release
Compiling MITIE as a shared library
On a UNIX like system, this can be accomplished by running make in the top level MITIE folder or by running:
cd mitielib
make
This produces shared and static library files in the mitielib folder. Or you can use CMake to compile a shared library by typing:
cd mitielib
mkdir build
cd build
cmake ..
cmake --build . --config Release --target install
Either of these methods will create a MITIE shared library in the mitielib folder.
Compiling MITIE using OpenBLAS
If you compile MITIE using cmake then it will automatically find and use any optimized BLAS libraries on your machine. However, if you compile using regular make then you have to manually locate your BLAS libaries or DLIB will default to its built in, but slower, BLAS implementation. Therefore, to use OpenBLAS when compiling without cmake, locate libopenblas.a and libgfortran.a, then run make as follows:
cd mitielib
make BLAS_PATH=/path/to/openblas.a LIBGFORTRAN_PATH=/path/to/libfortran.a
Note that if your BLAS libraries are not in standard locations cmake will fail to find them. However, you can tell it what folder to look in by replacing cmake .. with a statement such as:
cmake -DCMAKE_LIBRARY_PATH=/home/me/place/i/put/blas/lib ..
Using MITIE from a Python 2.7 program
Once you have built the MITIE shared library, you can go to the examples/python folder and simply run any of the Python scripts. Each script is a tutorial explaining some aspect of MITIE: named entity recognition and relation extraction, training a custom NER tool, or training a custom relation extractor.
You can also install mitie direcly from github with this command: pip install git+https://github.com/mit-nlp/MITIE.git.
Using MITIE from R
MITIE can be installed as an R package. See the README for more details.
Using MITIE from a C program
There are example C programs in the examples/C folder. To compile of them you simply go into those folders and run make. Or use CMake like so:
cd examples/C/ner
mkdir build
cd build
cmake ..
cmake --build . --config Release
Using MITIE from a C++ program
There are example C++ programs in the examples/cpp folder. To compile any of them you simply go into those folders and run make. Or use CMake like so:
cd examples/cpp/ner
mkdir build
cd build
cmake ..
cmake --build . --config Release
Using MITIE from a Java program
There is an example Java program in the examples/java folder. Before you can run it you must compile MITIE's java interface which you can do like so:
cd mitielib/java
mkdir build
cd build
cmake ..
cmake --build . --config Release --target install
That will place a javamitie shared library and jar file into the mitielib folder. Once you have those two files you can run the example program in examples/java by running run_ner.bat if you are on Windows or run_ner.sh if you are on a POSIX system like Linux or OS X.
Also note that you must have Swig 1.3.40 or newer, CMake 2.8.4 or newer, and the Java JDK installed to compile the MITIE interface. Finally, note that if you are using 64bit Java on Windows then you will need to use a command like:
cmake -G "Visual Studio 10 Win64" ..
instead of cmake .. so that Visual Studio knows to make a 64bit library.
Running MITIE's unit tests
You can run a simple regression test to validate your build. Do this by running the following command from the top level MITIE folder:
make test
make test builds both the example programs and downloads required example models. If you require a non-standard C++ compiler, change CC in examples/C/makefile and in tools/ner_stream/makefile.
Precompiled Python 2.7 binaries
We have built Python 2.7 binaries packaged with sample models for 64bit Linux and Windows (both 32 and 64 bit version of Python). You can download the precompiled package here: Precompiled MITIE 0.2
Precompiled Java 64bit binaries
We have built Java binaries for the 64bit JVM which work on Linux and Windows. You can download the precompiled package here: Precompiled Java MITIE 0.3. In the file is an examples/java folder. You can run the example by executing the provided .bat or .sh file.
License
MITIE is licensed under the Boost Software License - Version 1.0 - August 17th, 2003.
Permission is hereby granted, free of charge, to any person or organization obtaining a copy of the software and accompanying documentation covered by this license (the "Software") to use, reproduce, display, distribute, execute, and transmit the Software, and to prepare derivative works of the Software, and to permit third-parties to whom the Software is furnished to do so, all subject to the following:
The copyright notices in the Software and this entire statement, including the above license grant, this restriction and the following disclaimer, must be included in all copies of the Software, in whole or in part, and all derivative works of the Software, unless such copies or derivative works are solely in the form of machine-executable object code generated by a source language processor.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
References
[1] Davis E. King. Dlib-ml: A Machine Learning Toolkit. Journal of Machine Learning Research 10, pp. 1755-1758, 2009.
[2] Paramveer Dhillon, Dean Foster and Lyle Ungar, Eigenwords: Spectral Word Embeddings, Journal of Machine Learning Research (JMLR), 16, 2015.
[3] T. Joachims, T. Finley, Chun-Nam Yu, Cutting-Plane Training of Structural SVMs, Machine Learning, 77(1):27-59, 2009.
