A simple kbqa based on knowledge graph embedding
node and relation extracted via keyword based entity recognition multi relations extracted via part of speech with spacy graph search use pretrained graph embedding via rotatE model
-
Usage:
-
Train knowledge graph embedding
git clone [email protected]:zhupengjia/KnowledgeGraphEmbedding.git cd KnowledgeGraphEmbedding/codes ./run2.py --data_path=$DATA_PATH --model=RotatE --save_path=$SAVE_PATH --double_entity_embedding --cuda --hidden_dim=50
-
Run
./interact.py --checkpoint=$CHECKPOINTFILE --w2v_word2idx=$WORDEMBEDDINGLOOKUPFILE --w2v_idx2vec=$WORDEMBEDDINGWEIGHTFILE --backend restful
word2vec lookup and weight file can be generated from script in example directory:
cd example ./bert_to_wordvec.py
-