representation learning on networks
play

Representation Learning on Networks Yuxiao Dong Microsoft Research, - PowerPoint PPT Presentation

Representation Learning on Networks Yuxiao Dong Microsoft Research, Redmond Joint work with Jiezhong Qiu, Jie Zhang, Jie Tang (Tsinghua University) Hao Ma (MSR & Facebook AI) and Kuansan Wang (MSR) Networks Social networks Economic


  1. Representation Learning on Networks Yuxiao Dong Microsoft Research, Redmond Joint work with Jiezhong Qiu, Jie Zhang, Jie Tang (Tsinghua University) Hao Ma (MSR & Facebook AI) and Kuansan Wang (MSR)

  2. Networks Social networks Economic networks Biomedical networks Information networks Internet Networks of neurons Slides credit: Jure Leskovec

  3. The Network & Graph Mining Paradigm 𝑦 π‘—π‘˜ : node 𝑀 𝑗 ’s π‘˜ π‘’β„Ž feature, e.g., 𝑀 𝑗 ’s pagerank value Graph & network applications β€’ Node label inference; β€’ X Link prediction; β€’ User behavior… … hand-crafted feature matrix machine learning models feature engineering

  4. Representation Learning for Networks Graph & network applications β€’ Node label inference; β€’ Z Node clustering; β€’ Link prediction; β€’ … … hand-crafted latent feature matrix machine learning models Feature engineering learning β€’ Input: a network 𝐻 = (π‘Š, 𝐹) Output: 𝒂 ∈ 𝑆 π‘Š ×𝑙 , 𝑙 β‰ͺ |π‘Š| , 𝑙 -dim vector 𝒂 𝑀 for each node v . β€’

  5. Network Embedding: Random Walk + Skip-Gram π‘₯ π‘—βˆ’2 π‘₯ π‘—βˆ’1 π‘₯ 𝑗 π‘₯ 𝑗+1 π‘₯ 𝑗+2 β€’ sentences in NLP skip-gram β€’ vertex-paths in Networks (word2vec) Perozzi et al. DeepWalk: Online learning of social representations. In KDD’ 14 , pp. 701 – 710.

  6. Random Walk Strategies β€’ Random Walk – DeepWalk (walk length > 1) – LINE (walk length = 1) β€’ Biased Random Walk β€’ 2 nd order Random Walk – node2vec β€’ Metapath guided Random Walk – metapath2vec

  7. Application: Embedding Heterogeneous Academic Graph metapath2vec Microsoft Academic Graph β€’ https://academic.microsoft.com/ β€’ https://www.openacademic.ai/oag/ β€’ metapath2vec: scalable representation learning for heterogeneous networks. In KDD 2017.

  8. Application 1: Related Venues β€’ https://academic.microsoft.com/ β€’ https://www.openacademic.ai/oag/ β€’ metapath2vec: scalable representation learning for heterogeneous networks. In KDD 2017.

  9. Application 2: Similarity Search (Institution) Microsoft Facebook Stanford Harvard Johns Hopkins UChicago AT&T Labs Google MIT Yale Columbia CMU β€’ https://academic.microsoft.com/ β€’ https://www.openacademic.ai/oag/ β€’ metapath2vec: scalable representation learning for heterogeneous networks. In KDD 2017.

  10. Network Embedding Random Walk Skip Gram Output: Input: Vectors Adjacency Matrix 𝒂 𝑩 β€’ Random Walk – DeepWalk (walk length > 1) – LINE (walk length = 1) β€’ Biased Random Walk 2 nd order Random Walk β€’ – node2vec β€’ Metapath guided Random Walk – metapath2vec

  11. Unifying DeepWalk, LINE, PTE, & node2vec as Matrix Factorization β€’ DeepWalk β€’ LINE β€’ PTE β€’ node2vec 𝑩 Adjacency matrix b : #negative samples T : context window size 𝑬 Degree matrix π‘€π‘π‘š 𝐻 = ෍ 𝐡 π‘—π‘˜ ෍ 𝑗 π‘˜ 1. Qiu et al. Network embedding as matrix factorization: unifying deepwalk, line, pte, and node2vec. In WSDM’18.

  12. Understanding Random Walk + Skip Gram π‘₯ π‘—βˆ’2 π‘₯ π‘—βˆ’1 π‘₯ 𝑗 π‘₯ 𝑗+1 π‘₯ 𝑗+2 𝐻 = (π‘Š, 𝐹) β€’ (π‘₯, 𝑑) : co-occurrence of w & c ? log(#(𝒙, 𝒅)|𝒠| β€’ (π‘₯) : occurrence of node w β€’ Adjacency matrix 𝑩 𝑐#(π‘₯)#(𝑑)) β€’ (𝑑) : occurrence of context c β€’ Degree matrix 𝑬 β€’ 𝒠: nodeβˆ’context pair (w, c) multiβˆ’set β€’ Volume of 𝐻: π‘€π‘π‘š 𝐻 β€’ |𝒠| : number of node-context pairs Levy and Goldberg. Neural word embeddings as implicit matrix factorization. In NIPS 2014

  13. Understanding Random Walk + Skip Gram log(#(𝒙, 𝒅)|𝒠| 𝑐#(π‘₯)#(𝑑)) β€’ (π‘₯, 𝑑) : co-occurrence of w & c β€’ (π‘₯) : occurrence of node w β€’ (𝑑) : occurrence of context c β€’ 𝒠: nodeβˆ’context pair (w, c) multiβˆ’set β€’ |𝒠| : number of node-context pairs

  14. Understanding Random Walk + Skip Gram log(#(𝒙, 𝒅)|𝒠| 𝑐#(π‘₯)#(𝑑)) β€’ (π‘₯, 𝑑) : co-occurrence of w & c β€’ (π‘₯) : occurrence of node w β€’ (𝑑) : occurrence of context c β€’ 𝒠: nodeβˆ’context pair (w, c) multiβˆ’set β€’ |𝒠| : number of node-context pairs β€’ Partition the multiset 𝒠 into several sub-multisets according to the way in which each node and its context appear in a random walk node sequence. β€’ More formally, for 𝑠 = 1, 2, β‹― , π‘ˆ , we define Distinguish direction and distance

  15. Understanding Random Walk + Skip Gram the length of random walk 𝑀 β†’ ∞ β€’ (π‘₯, 𝑑) : co-occurrence of w & c β€’ 𝒠: (w, c) multiβˆ’set

  16. Understanding Random Walk + Skip Gram the length of random walk 𝑀 β†’ ∞

  17. Understanding Random Walk + Skip Gram π‘₯ π‘—βˆ’2 π‘₯ π‘—βˆ’1 π‘₯ 𝑗 π‘₯ 𝑗+1 π‘₯ 𝑗+2 DeepWalk is asymptotically and implicitly factorizing 𝑩 Adjacency matrix 𝑬 Degree matrix π‘€π‘π‘š 𝐻 = ෍ ෍ 𝐡 π‘—π‘˜ 𝑗 π‘˜ b : #negative samples T : context window size Qiu et al. Network embedding as matrix factorization: unifying deepwalk, line, pte, and node2vec. In WSDM’18. 1.

  18. Unifying DeepWalk, LINE, PTE, & node2vec as Matrix Factorization β€’ DeepWalk β€’ LINE β€’ PTE β€’ node2vec Qiu et al. Network embedding as matrix factorization: unifying deepwalk, line, pte, and node2vec. In WSDM’18. The most cited paper in WSDM’18 as of May 2019

  19. NetMF: explicitly factorizing the DeepWalk matrix Matrix π‘₯ π‘—βˆ’2 π‘₯ π‘—βˆ’1 π‘₯ 𝑗 Factorization π‘₯ 𝑗+1 π‘₯ 𝑗+2 DeepWalk is asymptotically and implicitly factorizing Qiu et al. Network embedding as matrix factorization: unifying deepwalk, line, pte, and node2vec. In WSDM’18. 1.

  20. the NetMF algorithm 1. Construction 2. Factorization 𝑻 = Qiu et al. Network embedding as matrix factorization: unifying deepwalk, line, pte, and node2vec. In WSDM’18. 1.

  21. Results β€’ Predictive performance on varying the ratio of training data; β€’ The x -axis represents the ratio of labeled data (%) Qiu et al. Network embedding as matrix factorization: unifying deepwalk, line, pte, and node2vec. In WSDM’18. 1.

  22. Results Explicit matrix factorization (NetMF) offers performance gains over implicit matrix factorization (DeepWalk & LINE) Qiu et al. Network embedding as matrix factorization: unifying deepwalk, line, pte, and node2vec. In WSDM’18. 1.

  23. Network Embedding Random Walk Skip Gram DeepWalk, LINE, node2vec, metapath2vec (dense) Matrix Output: 𝑻 = 𝑔(𝑩) Input: Factorization Vectors Adjacency Matrix NetMF 𝒂 𝑩 Incorporate network structures 𝑩 into the similarity matrix 𝑻 , and then factorize 𝑻 𝑔 𝑩 =

  24. Challenges 𝑻 = NetMF is not practical for very large networks

  25. NetMF How can we solve this issue? 1. Construction 2. Factorization 𝑻 = 1. Qiu et al. NetSMF: Network embedding as sparse matrix factorization. In WWW 2019

  26. NetSMF--Sparse How can we solve this issue? 1. Sparse Construction 2. Sparse Factorization 𝑻 = 1. Qiu et al. NetSMF: Network embedding as sparse matrix factorization. In WWW 2019

  27. Sparsify 𝑻 For random-walk matrix polynomial where and non-negative One can construct a 1 + πœ— -spectral sparsifier ΰ·¨ 𝑴 with non-zeros in time for undirected graphs β€’ Dehua Cheng, Yu Cheng, Yan Liu, Richard Peng, and Shang-Hua Teng, Efficient Sampling for Gaussian Graphical Models via Spectral Sparsification, COLT 2015. β€’ Dehua Cheng, Yu Cheng, Yan Liu, Richard Peng, and Shang-Hua Teng. Spectral sparsification of random-walk matrix polynomials. arXiv:1502.03496.

  28. Sparsify 𝑻 For random-walk matrix polynomial where and non-negative One can construct a 1 + πœ— -spectral sparsifier ΰ·¨ 𝑴 with non-zeros in time 𝑻 = 1. Qiu et al. NetSMF: Network embedding as sparse matrix factorization. In WWW 2019

  29. NetSMF --- Sparse Factorize the constructed sparse matrix 1. Qiu et al. NetSMF: Network embedding as sparse matrix factorization. In WWW 2019

  30. NetSMF---bounded approximation error 𝑡 ΰ·© 𝑡 1. Qiu et al. NetSMF: Network embedding as sparse matrix factorization. In WWW 2019

  31. #non-zeros ~4.5 Quadrillion β†’ 45 Billion 1. Qiu et al. NetSMF: Network embedding as sparse matrix factorization. In WWW 2019

  32. 1. Qiu et al. NetSMF: Network embedding as sparse matrix factorization. In WWW 2019

  33. Effectiveness: β€’ (sparse MF)NetSMF β‰ˆ (explicit MF)NetMF > (implicit MF) DeepWalk/LINE Efficiency: β€’ Sparse MF can handle billion-scale network embedding 1. Qiu et al. NetSMF: Network embedding as sparse matrix factorization. In WWW 2019

  34. Embedding Dimension? 1. Qiu et al. NetSMF: Network embedding as sparse matrix factorization. In WWW 2019

  35. Network Embedding Random Walk Skip Gram DeepWalk, LINE, node2vec, metapath2vec (dense) Matrix Output: 𝑻 = 𝑔(𝑩) Input: Factorization Vectors Adjacency Matrix NetMF 𝒂 𝑩 (sparse) Matrix Sparsify 𝑻 Factorization NetSMF Incorporate network structures 𝑩 into the similarity matrix 𝑻 , and then factorize 𝑻 𝑔 𝑩 =

  36. ProNE: More fast & scalable network embedding 1. Zhang et al. ProNE: Fast and Scalable Network Representation Learning. In IJCAI 2019

Recommend


More recommend