Curvature Regularization to Prevent Distortion in Graph Embedding

Authors: Hongbin Pei, Bingzhe Wei, Kevin Chang, Chunxu Zhang, Bo Yang

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We integrate curvature regularization into five popular proximity-preserving embedding methods, and empirical results in two applications show significant improvements on a wide range of open graph datasets.
Researcher Affiliation Academia 1College of Computer Science and Technology, Jilin University, China 2Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, USA 3Department of Computer Science, University of Illinois at Urbana-Champaign, USA 4Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education, China
Pseudocode Yes Algorithm 1 Curvature regularization optimizing algorithm. 1: input: graph G 2: output: node representations X 3: preprocessing: get paths in G Shortest paths (Ωc and Ωs) or random-walk paths (Ωa) 4: for t iterations do First phase 5: while not converged do 6: minimize embedding loss term L(G) 7: while not converged do 8: minimize curvature regularization term Ω(X) Specific term is Ωc, Ωs, or Ωa 9: while not converged do Second phase 10: minimize the two terms jointly L(G) + λΩ(X) λ is a trade-off hyperparameter
Open Source Code No The paper does not provide an explicit statement or link for the release of its source code.
Open Datasets Yes Datasets. We evaluate the proposed method on eight open graph datasets described below (more details are available in the Appendix). (1) Citation networks. Cora, Citeseer and Pubmed are citation network benchmark datasets [22, 23], where nodes are papers and edges are citation links. ... (2) Web KB. ... (3) Polblogs. Political blog network [24]...
Dataset Splits Yes we split randomly 60% of the nodes in a graph as the training set and the remaining 40% of nodes in a graph as the test set. ... For all node embedding models, we perform a random sampling hyper-parameter search on validation set of each dataset to get competitor models (See detailed hyper-parameter setting in the Appendix).
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes Parameter search. For all node embedding models, we perform a random sampling hyper-parameter search on validation set of each dataset to get competitor models (See detailed hyper-parameter setting in the Appendix). The hyper-parameters searched over include the dimension of node representation as well as hyper-parameters specific to each model. We then integrate the curvature regularization term into those competitors, and only adjust the number of iteration t and the weight λ in algorithm.