iamaziz / kge_qa

Simple-Question Answering based on Knowledge Graph Embeddings (part of my Ph.D. work)

KGE QA

Simple-Questions Answering System based on Knowledge Graphs Embeddings.

Setup

$ virtualenv .env -p python3.7
$ source .env/bin/activate
$ pip install -r requirements.txt

Getting Started

Run in command line mode:

$ python -m kgeqa.main

Run in web browser mode - requires streamlit:

$ streamlit run app.py

Building KGE for a new domain knowledge

We can build new KGE for a new KG dataset either from the CLI or UI (streamlit interface)

From command-line:

$ python -m kgeqa.build_new_model -csv data/sample1_KG.csv
Started a model builder for data from: data/sample1_KG.csv
Building a new embedding model for 15 tokens ..
Done. See output: data/ENT.vec
Building a new embedding model for 7 tokens ..
Done. See output: data/REL.vec
Converting models to .magnitude format ..
Loading vectors... (this may take some time)
Found 15 key(s)
Each vector has 300 dimension(s)
Creating magnitude format...
Writing vectors... (this may take some time)
...
Successfully converted 'data/ENT.vec' to 'data/ENT.vec.magnitude'!
...
Successfully converted 'data/REL.vec' to 'data/REL.vec.magnitude'!
Done.

Description of the generated models:

• data/ENT.vec Entity model in .txt format (intermediate result - not used in the app)
• data/ENT.vec.magnitude Entity model in PyMagnitude format
• data/REL.vec Relation model in .txt format (intermediate result - not used in the app)
• data/REL.vec.magnitude Relation model in PyMagnitude format

Author

• Aziz Altowayan (Nov. 2019)