Text2KG

UWA's submission to the ICDM2019 ICDM/ICBK Knowledge Graph Contest was selected by the committee as the first prize.

Team members:

• Wei Liu
• Majigsuren Enkhsaikhan
• Michael Stewart
• Morgan Lewis
• Thomas Smoker

File structure

datasets // Contains all contest-provided datasets
sourcecode
    candidate_extraction    // Coref resolution + identifying noun and relation phrases
    visualisation           // Visualisation web app
    requirements.txt        // All Python requirements for the /sourcecode directory
    setup.sh                // Creates a virtual environment and installs required packages
    run.sh                  // Runs the triple generation code

Running the code

The sourcecode, available under the /sourcecode directory, may be run via the following commands:

cd sourcecode
chmod u+x setup.sh
chmod u+x run.sh
./setup.sh
./run.sh

Please ensure you are using Python 3.6 and have the virtualenv package installed in order to maximise compatibility. Running the shell scripts will likely require a Unix-based OS such as Ubuntu.

Visualisation web app

We have developed a Flask application that provides an easy-to-use interface for knowledge graph construction from text. Users may enter documents into a text box, click on the "Create graph" button, and their documents will be sent to our Candidate Extraction model for processing. An interactive graph will then be quickly be drawn to the screen.

If you wish to run the server locally, you must run the Visualisation server via:

$ cd sourcecode/visualisation
$ chmod u+x run_server.sh
$ ./run_server.sh

The visualisation app will then be available at localhost:5000 in a web browser.

You can also run ./run_server_gunicorn.sh if you wish to run the server via wsgi.

Running the visualisation with a different triple generation backend

It is possible to run the visualisation app with a different model for generating the triples. In line 39 of sourcecode/visualisation/app/routes.py, simply replace the extract_triples function with a different function that returns a list of triples, given a document. For example, if given the document "Barrack Obama went to the Whitehouse with Michelle", the function must take that string as input and return a list of triples, e.g. [["Barrack", "went to", "the Whitehouse"], ["Barrack Obama", "with", "Michelle"]... ].

License

text2kg-uwa (c) by Michael Stewart, Majigsuren Enkhsaikhan and Wei Liu

text2kg-uwa is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

Reference

@inproceedings{stewart2019icdm,    
    title={ICDM 2019 Knowledge Graph Contest: Team UWA},
    author={Stewart, Michael and Enkhsaikhan, Majigsuren and Liu, Wei},
    booktitle={ICDM 2019},
    year={2019}
}