Closed alanyuchenhou closed 7 years ago
The old model R implementation used many deprecated tensorflow modules so I did a rewrite. The new implementation has better performance in forum data set: However, the performance gain is not significant enough to make a point like "Model R outperforms node2vec in all scenarios".
True. But looks like Model R should be preferred to node2vec. Model R probably significantly outperforms node2vec in Collaboration graph. And Model R does not worse than node2vec on all domains.
I adopted LLE from all GEM approaches.
Found a package that implements a few graph and node embedding methods: https://github.com/palash1992/GEM