The implementation of HDSKG using BERT model

Paper link

Paper link

Preface😘

• Limited to my ability, there is still much room for improvement. 😂
• I would appreciate it if you could give me some advice on this work. 😄

The architecture of HDSKG is as follows:

Installation

1. pip install -r requirements.txt
2. I use stanza to access Stanford CoreNLP, I suggest installing it from source of its git repository

   git clone https://github.com/Rvlis/stanza.git
   cd stanza
   pip install -e .

3. Manually download Stanford CoreNLP or use stanza.install_corenlp("your/absolute/path")
4. Setting the CORENLP_HOME environment variable with the your/absolute/path
5. I use Spacy and neuralcoref to resolve coreference, install neuralcoref from source as well

   git clone https://github.com/Rvlis/neuralcoref.git
   cd neuralcoref
   pip install -r requirements.txt
   pip install -e .

6. Don't forget loading spacy's model, which is dependent on your spacy's version. Here 2.3.* is ok.

Usage

1. HDSKG-Chunking
2. HDSKG-Domain-Relevance-Estimation
3. Generate-Domain-Knowledge-Graph

Test Bert Model

Compared with previous 0.76, accuracy is improved to 0.82 now. The checkpoint file will be uploaded soon.

Knowledge Graph

TODO

• package all parts
• cluster all relations

Keep Going!🐱‍🏍