Out-of-knowledge-base entities linking

1. Introduction:

This repository is the code part of the CMPUT 692 class project: A study of out-of-KB entities.

2. Dependencies:

For finding OOKB entities from text

pip install spacy
python -m spacy download en_core_web_md

For training and testing model

pip install -U sentence-transformers
pip install torch
pip install numpy
pip install tqdm

3. Dataset & model download:

After downloading, unzip and put in the corresponding directory.

• openKE_scripts/dbpedia50_openKE
• open_world/dbpedia50
• open_world/openke_models

4. Execution:

Recognize out-of-KB entities in a document

cd open_world
python OOKB_entities_recognition.py

The program will ask for a file name. A sample document is in open_world/example_text/sample.txt

Training a closed-world model:

Thanks to OpenKE KGC framework

cd openKE_scripts/openke
bash make.sh
cd ..
python train_transe300D_DBpedia50k.py

Traning a open-world model:

cd open-world
python run_open_world.py --mode train

Benchmarking:

cd open-world
python run_open_world.py --mode benchmark

Predict a single OOKB entity with the description:

Prepare a file with the following format:
First line: ID of the relation
Second line: description of the entity

cd open-world
python run_open_world.py --mode predict --file [filename]

A sample file is in open_world/example_text/test_predict.txt

5. Acknowledge:

• Training the closed-world model using OpenKE (https://github.com/thunlp/OpenKE)
• Inspired by the OWE model (https://github.com/haseebs/OWE)