This is the implementation of the Hierarchical Representation Learning on Heterogeneous Graph(HRLHG) model from paper "Cross-language Citation Recommendation via Hierarchical Representation Learning on Heterogeneous Graph". Note that this is only a reference implementation of the HRLHG algorithm and could benefit from several performance enhancement schemes, some of which are discussed in the paper.

numpy
networkx
gensim

1.generate Relation Type Usefulness Distributions (RTUD) via K-shortest paths ranking based EM algorithm
2.generate node sequence via hierarchical (two-level) random walk algorithm based on RTUD, and generate node embedding
3.generate ranking result according to the cosine similarity between nodes
4.evaluate node embedding result via IR metrics

node embedding needs 2 days in normal CPU machine with follow settings:
walk-length: 80
dimension: 128
num-walks: 10
window-size: 10

The time complexity can be further reduced, if we parallelize the hierarchical random walk simulations, and execute them asynchronously

Zhuoren Jiang, Yue Yin, Liangcai Gao, Yao Lu, and Xiaozhong Liu. 2018. 
Cross-language Citation Recommendation via Hierarchical Representation Learning on Heterogeneous Graph. 
In SIGIR ’18: The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval, July 8–12, 2018, Ann Arbor, MI, USA. 
ACM, New York, NY, USA, 10 pages. https://doi.org/10.1145/3209978.3210032

@inproceedings{jiang2018cross,
  title={Cross-language Citation Recommendation via Hierarchical Representation Learning on Heterogeneous Graph},
  author={Jiang, Zhuoren and Yin, Yue and Gao, Liangcai and Lu, Yao and Liu, Xiaozhong},
  booktitle={The 41st International ACM SIGIR Conference on Research \& Development in Information Retrieval},
  pages={635--644},
  year={2018},
  organization={ACM}
}