A semisupervised learning model for heterogeneous information networks (HINs)
ujson == 1.35 pytorch >= 1.5.0
-
(DBLP dataset only) Download Glove word embdding: http://nlp.stanford.edu/data/glove.840B.300d.zip
-
Generate node embeddings by node2vec or metapath2vec: a. preprocess/yelp.py -> gen_homograph() preprocess/main.py --input ../../data/yelp/homograph.txt --output ../../data/ yelp/RUBK_128.emb --dimensions 128 --workers 56 --walk-length 100 --num-walks 40 --window-size 5 b. preprocess/yelp.py -> gen_walk(path='../data/yelp/', walk_length=100,n_walks=40) preprocess/metapath2vec -train ../../data/yelp/BRKRB.walk -output ../../data/yelp/BRKRB_128.emb -size 128 -threads 40
-
Fuse edge features: preprocess/yelp.py -> dump_yelp_edge_emb(path='../data/yelp/')
-
Compute index: preprocess/yelp.py -> gen_edge_adj_random(path='../data/yelp/',edge_dim=130)
run_dblp.py; run_freebase.py; run_yelp.py
Conch: train_reg3.py Conch_nc: train_ncreg.py Conch_rd: train_rdreg.py
Conch: train.py Conch_nc: train_nc.py Conch_rd: train_rd.py
MIT