yangkeyin / mshine

The official Tensorflow implementation of mSHINE: A Multiple-meta-paths Simultaneous Learning Framework for Heterogeneous Information Network Embedding.

mSHINE

The official Tensorflow implementation of mSHINE: A Multiple-meta-paths Simultaneous Learning Framework for Heterogeneous Information Network Embedding. mSHINE.

---

Package Version

 Keras-Preprocessing==1.1.2
 numpy==1.19.2
 scikit-learn==0.24.1
 scipy==1.4.1
 tensorflow-gpu==2.2.0

---

Data Preparation

The experimental datasets used in paper are available at https://drive.google.com/file/d/1g3Ln0fzCIqUO7A1GTZpKntXSUaiJdqgZ/view?usp=sharing

To run mSHINE on your HIN whose name is XXX(we use the dataset cora as an example blow), two files should be provided: 1)XXX.hin (HIN dataset); 2)XXX.config (Meta-path info)

1. The supported input HIN format is an edgelist (separated by space):

    node_type_1:node_id_1 node_type_2:node_id_2 edge_weights node_type_1:node_type_2
    ...
   

2. Generate XXX.config:

   • XXX.relation file which lists all the possible egde types should be provided. The format of XXX.relation:

    node_type_1-node_type_2
    node_type_2-node_type_1
    node_type_1-node_type_3
    ...
   

   (NOTE: The edge is assumed to be directed by default, you need to use two directed edges to represent an undirected edge type.)

   • To generate XXX.config:

    python config_file_gen/metapath_gen.py --dn cora --output_dir config_file_gen/config_files/ --relation_dir config_file_gen/relation_files/
   

3. Generate other necessary dataset files(i.e. XXX.HIN_dic, XXX_id_to_index.p and XXX_info_dict.p):

   • Be sure XXX.hin and XXX.config are put in the folder: data/XXX/, then:

    cd data_prepare/
    python data_prepare.py --input_data cora

   The resulted files can be found in data/XXX/

---

Excute

To run mSHINE:

 python main.py --graph_name cora --dimensions 128 --batch_size 128 --iter 1000

---

Output

• The node representaions XXX_{epoch}_{emb_size}.emb can be found in: data/XXX/experi_data/node_emb/

  the embeddings are stored in the form of dictionary in a pickle file and can be loaded:

   import pickle
     
   emb = pickle.load(open('{path_of_emb_file}'),'rb'))
   state_embedding = emb['s_embedding']
   input_embedding = emb['i_embedding']
   output_embedding = emb['o_embedding']
  

  (NOTE: the mapping from index of node in state_embedding matrix to the node id can be found in XXX_id_to_index.p )

• The transform matrix stored in trans_metric_{epoch}_{emb_size} can be found in: data/XXX/experi_data/record/

---

Evaluation

• An example of evalution code is available in node_emb_classification.py where a label file XXX_label.txt is required.

  The format of XXX_label.txt is:

   node_id_1 node_label
   node_id_2 node_label
   ...
  

  To run the evaluation:

    python node_emb_classification.py --dimensions 128 --experi_dir experi_data --target_node_type p --graph_name cora
  

  during which, XXX_TEST.label is generated for storing classification related info.

  (NOTE: the result reported in the paper is evaluated through SVM with kernel='rbf', gamma='scale').

---

Citing

If you find mSHINE is useful for your research, please consider citing the following paper:

@ARTICLE{9201301,  
    author={X. {Zhang} and L. {Chen}},  
    journal={IEEE Transactions on Knowledge and Data Engineering},   
    title={mSHINE: A Multiple-meta-paths Simultaneous Learning Framework for Heterogeneous Information Network Embedding},   
    year={2020},
    volume={},
    number={},
    pages={1-1},
    doi={10.1109/TKDE.2020.3025464}}

Please send any questions you might have about the codes and/or the algorithm to [email protected].