IP Library › Granted Patent US 12,554,997
Granted Patent B2
US 12,554,997 · App. 17/154,123 · Granted Feb 17, 2026

Deep multi-view network embedding on incomplete data

Inventors: Qifan Wang (Sunnyvale, CA); Ruining He (Mountain View, CA)
Assignee: GOOGLE LLC
G06N5/022G06F17/16G06N3/02G06N20/00H04L41/12
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,554,997
App. No.
17/154,123
Granted
Feb 17, 2026
Kind
B2
Abstract

A computing system can obtain a plurality of datasets that respectively correspond to a plurality of views of a graph network comprising a plurality of nodes, wherein the plurality of datasets comprise one or more partial view datasets for one or more partial views that comprise data for only a respective subset of the plurality of nodes of the graph network. The computing system can determine, based at least in part on an objective function, a plurality of respective embeddings associated respectively with the plurality of nodes, such as a common embedding set that contains respective embeddings for a common subset of the plurality of nodes that are common among all of the plurality of views, and one or more independent embedding sets that contain respective embeddings for one or more respective subsets of the plurality of nodes described only by a subset of the plurality of views.

Claims (54)

1 . A computer-implemented method to generate embeddings from partial view data, the method comprising:

obtaining, by one or more computing devices, a plurality of datasets that respectively correspond to a plurality of views of a graph network comprising a plurality of nodes, wherein the plurality of datasets comprise one or more partial view datasets for one or more partial views that comprise data for only a respective subset of the plurality of nodes of the graph network;

initializing, by the one or more computing devices:

a common embedding set that contains respective embeddings for a common subset of the plurality of nodes that are common among all of the plurality of views; and

one or more independent embedding sets that contain respective embeddings for one or more respective independent subsets of the plurality of nodes that are described by only a respective subset of the plurality of views;

initializing, by the one or more computing devices for each respective view of the plurality of views, a view-specific basis matrix of a plurality of view-specific basis matrices and a view-specific encoder model of a plurality of view-specific encoder models; and

updating the plurality of view-specific encoder models, the plurality of view-specific basis matrices, and the one or more independent embedding sets using a coordinate descent operation, wherein the coordinate descent operation comprises, for each of a plurality of iterations:

processing, by the one or more computing devices, each of the plurality of datasets with a corresponding view-specific encoder model of the plurality of view-specific encoder models to obtain, for each of the plurality of datasets, respective encoded representations of the plurality of nodes described by the corresponding view;

updating, by the one or more computing devices for each respective view of the plurality of views, based at least in part on an objective function, the view-specific encoder model;

updating, by the one or more computing devices for each respective view of the plurality of views, based at least in part on the objective function, the view-specific basis matrix;

updating, by the one or more computing devices, based at least in part on the objective function, the common embedding set; and

updating, by the one or more computing devices, based at least in part on the objective function, the one or more respective independent embedding sets;

wherein the objective function comprises a comparison of the respective encoded representations with the respective embeddings, and wherein the objective function further comprises a proximity preservation term that enforces preservation of proximity within each view.

2 . The computer-implemented method of claim 1 , wherein the one or more independent embedding sets comprise a plurality of independent embedding sets.

3 . The computer-implemented method of claim 1 , wherein at least one of the one or more independent embedding sets contains respective embeddings for a respective independent subset of the plurality of nodes that is described by only a single view.

4 . The computer-implemented method of claim 1 , wherein at least one of the one or more independent embedding sets contains respective embeddings for a respective independent subset of the plurality of nodes that is described by two or more but not all of the plurality of views.

5 . The computer-implemented method of claim 1 , wherein, for at least one of the plurality of views, the comparison is performed using a respective embedding matrix that comprises separate respective entries for the common embedding set and at least one of the one or more independent embedding sets and wherein, for the at least one of the plurality of views, the respective embedding matrix comprises a first entry for the common embedding set and a second entry for a respective independent subset of the plurality of nodes that are described by only such view.

6 . The computer-implemented method of claim 1 , wherein updating the view-specific encoder comprises:

modifying, by the one or more computing devices, one or more parameter values of the view-specific encoder model to reduce the objective function.

7 . The computer-implemented method of claim 1 , wherein said updating the common embedding set comprises minimizing the objective function.

8 . The computer-implemented method of claim 1 , further comprising analyzing one or more of the plurality of nodes of the graph network based on the common embedding set and the one or more independent embedding sets.

9 . The computer-implemented method of claim 1 , wherein the objective function evaluates, for each of the plurality of views, a respective difference between the respective encoded representations of the plurality of nodes obtained for such view and the respective embeddings for the nodes described by such view multiplied by a respective basis matrix for such view.

10 . The computer-implemented method of claim 1 , wherein the objective function evaluates, for each of the plurality of views, a Frobenius norm of a respective difference between the respective encoded representations of the plurality of nodes obtained for such view and the respective embeddings for the nodes described by such view multiplied by a respective basis matrix for such view.

11 . The computer-implemented method of claim 1 , wherein the objective function comprises a sum of respective distances evaluated for the plurality of views.

12 . The computer-implemented method of claim 1 , wherein at least two of the plurality of datasets comprise different respective modalities of data.

13 . A first computing system, comprising:

one or more processors; and

one or more non-transitory computer-readable media that collectively store instructions that when executed by the one or more processors cause the computing system to perform operations, the operations comprising:

obtaining a unified embedding for a graph network comprising a plurality of nodes, wherein the unified embedding has been generated from a plurality of datasets that respectively correspond to a plurality of views of the graph network, wherein the plurality of datasets comprise one or more partial view datasets for one or more partial views that comprise data for only a respective subset of the plurality of nodes of the graph network, and wherein the unified embedding comprises:

a common embedding set that contains respective embeddings for common subset of the plurality of nodes that are common among all of plurality of views; and

one or more independent embedding sets that contain respective embeddings for one or more respective independent subsets of the plurality of nodes that are described by only a respective subset of the plurality of views; and

analyzing one or more of the plurality of nodes of the graph network based on the unified embedding;

wherein the unified embedding was generated by:

initializing, by the first computing system or a second computing system comprising one or more computing devices: the common embedding set; the one or more independent embedding sets; and, for each respective view of the plurality of views, a view-specific basis matrix of a plurality of view-specific basis matrices and a view-specific encoder model of a plurality of view-specific encoder models; and

updating the plurality of view-specific encoder models, the plurality of view-specific basis matrices, and the one or more independent embedding sets using a coordinate descent operation, wherein the coordinate descent operation comprises, for each of a plurality of iterations:

processing, by the first or second computing system, each of the plurality of datasets with a corresponding view-specific encoder model of the plurality of view-specific encoder models to obtain, for each of the plurality of datasets, respective encoded representations of the plurality of nodes described by the corresponding view;

updating, by the one or more computing devices for each respective view of the plurality of views, based at least in part on an objective function, the view-specific encoder model;

updating, by the one or more computing devices for each respective view of the plurality of views, based at least in part on the objective function, the view-specific basis matrix;

updating, by the one or more computing devices, based at least in part on the objective function, the common embedding set; and

updating, by the one or more computing devices, based at least in part on the objective function, the one or more respective independent embedding sets;

wherein the objective function comprises a comparison of the respective encoded representations with the respective embeddings, and wherein the objective function further comprises a proximity preservation term that enforces preservation of proximity within each view.

14 . The computing system of claim 13 , wherein the graph network comprises a social network that describes social connections between entities.

15 . The computing system of claim 13 , wherein the graph network comprises a logistics network that describes logistical connections between logistical nodes.

16 . The computing system of claim 13 , wherein the graph network comprises a biological network that describes biological units.

17 . The computing system of claim 13 , wherein the graph network comprises a chemical network that describes chemical units.

18 . The computing system of claim 13 , wherein analyzing one or more of the plurality of nodes based on the unified embedding comprises:

using the respective embedding for a first node of the plurality of nodes to:

identify one or more other nodes of the plurality of nodes that are similar to the first node;

classify the first node;

predict a link between the first node and at least one other node of the plurality of nodes;

cluster the first node with the one or more other nodes of the plurality of nodes; or

identify a community to which the first node belongs.

19 . The computing system of claim 13 , wherein the unified embedding was generated by the second computing system and wherein the second computing system is distinct from the first computing system.

20 . The computing system of claim 13 , wherein the unified embedding was generated by the first computing system.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jan 22, 2021
From: WANG, QIFAN; HE, RUINING
To: GOOGLE LLC
Reel/Frame 054992/0323 →
Continuity (2)
Provisional Application 62963756 · Jan 21, 2020
Related Publication 20210224352A1 · Jul 22, 2021
References Cited (47)
US 10997219B1 · Han · 2021 [cited by examiner]
US 11562226B2 · Saito · 2023 [cited by examiner]
US 20160246919A1 · Wang · 2016 [cited by examiner]
US 20190303857A1 · Lecue · 2019 [cited by examiner]
Wang et al., MGAE: Marginalized Graph Autoencoder for Graph Clustering, Nov. 6, 2017 (Year: 2017). [cited by examiner]
Zhang et al., Multi-view Knowledge Graph Embedding for Entity Alignment, Jun. 30, 2013 (Year: 2013). [cited by examiner]
Rai et al., Partial Multi-view Clustering Using Graph Regularized NMF, Apr. 24, 2016 (Year: 2016). [cited by examiner]
Xu et al., “Adversarial Incomplete Multi-view Clustering”, Aug. 10, 2019, IJCAI, vol. 7, pp. 3933-3939. (Year: 2019). [cited by examiner]
Wen et al., “Unified Embedding Alignment with Missing Views Inferring for Incomplete Multi-View Clustering”, Jul. 17, 2019, Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33 No. 01, pp. 5393-5400. (… [cited by examiner]
Ou et al., “Asymmetric Transitivity Preserving Graph Embedding”, Aug. 13, 2016, Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 1105-1114. (Year: 2016). [cited by examiner]
Ma et al., “Multi-view Clustering with Graph Embedding for Connectome Analysis”, Nov. 6, 2017, Proceedings of the 2017 ACM on Conference on Information and Knowledge Management, pp. 127-136. (Year: 2017). [cited by examiner]
Armandpour et al., “Robust Negative Sampling for Network Embedding”, 33rd AAAI Conference on Artificial Intelligence, Jan. 27-Feb. 1, 2019, Honolulu, Hawaii, pp. 3191-2198. [cited by applicant]
Belkin et al., “Laplacian Eigenmaps and Spectral Techniques for Embedding and Clustering”, Neural Information Processing Systems: Natural and Synthetic, Dec. 3-8, 2001, Vancouver, British Columbia, Canada, pp. 585-591. [cited by applicant]
Bhagat et al., “Node Classification in Social Networks”, arXiv:1101.3291v1, Jan. 17, 2011, 37 pages. [cited by applicant]
Bui et al., “Labeling Actors in Multi-view Social Networks by Integrating Information From Within and Across Multiple Views”, IEEE International Conference on Big Data, Dec. 5-8, 2016, Washington, D.C., 10 pages. [cited by applicant]
Cavallari et al., “Learning Community Embedding with Community Detection and Node Embedding on Graphs”, International Conference on Information and Knowledge Management (CIKM), Nov. 6-10, 2017, Pan Pacific, Singapore, p… [cited by applicant]
Du et al., “Galaxy Network Embedding: A Hierarchical Community Structure Preserving Approach”, International Joint Conference (IJCAI), Jul. 13-19, 2018, Sweden, Stockholm, pp. 2079-2085. [cited by applicant]
Gao et al., “Deep Attributed Network Embedding”, International Joint Conference (IJCAI), Jul. 13-19, 2018, Sweden, Stockholm, pp. 3364-3370. [cited by applicant]
Gao et al., “ProGAN: Network Embedding via Proximity Generative Adversarial Network”, 26 [cited by applicant]
Gong et al., “Iterative Quantization: A Procrustean Approach to Learning Binary Codes for Large-scale Image Retrieval”, IEEE Transaction on Pattern Analysis and Machine Intelligence, vol. 35, No. 12, 2013, 15 pages. [cited by applicant]
Grover et al., “node2vec: Scalable Feature Learning for Networks”, ACM SIGKDD Conference on Knowledge, Discovery and Data Mining, Aug. 13-17, 2016, San Francisco, CA, 10 pages. [cited by applicant]
Guo et al., “Anchors Bring Ease: An Embarrassingly Simple Approach to Partial Multi-View Clustering”, Thirty-Third AAAI Conference on Artificial Intelligence (AAAI-19), Jan. 27-Feb. 1, 2019, Honolulu, Hawaii, pp. 118-12… [cited by applicant]
Hamilton et al., “Inductive Representation Learning on Large Graphs”, 31 [cited by applicant]
Jin et al., “Incorporating Network Embedding into Markov Random Field for Better Community Detection”, Thirty Third AAAI Conference on Artificial Intelligence (AAAI-19), Jan. 27-Feb. 1, 2019, Honolulu, Hawaii, pp. 160-1… [cited by applicant]
Li et al., “Learning Network Embedding with Community Structural Information”, Twenty Eighth International Joint Conference on Artificial Intelligence (IJCAI-19), Aug. 10-16, 2019, Macao, China, pp. 2937-2943. [cited by applicant]
Li et al., “Partial Multi-View Clustering”, Twenty Eighth AAAI Conference on Artificial Intelligence, Jul. 27-31, 2014, Quebec City, Quebec, Canada, pp. 1968-1974. [cited by applicant]
Liben-Nowell et al., “The Link-Prediction Problem for Social Networks”, Journal of the American Society for Information Science and Technology, vol. 58, No. 7, May 2007. 23 pages. [cited by applicant]
Memisevic, “On Multi-view feature learning”, 29 [cited by applicant]
Perozzi et al., “DeepWalk: Online Learning of Social Representations” arXiv:1403.6652v2, Jun. 27, 2014, 10 pages. [cited by applicant]
Qu et al., “An Attention-based Collaboration Framework for Multi-View Network Representation Learning”, arXiv:1709.06636v1, Sep. 19, 2017, 10 pages. [cited by applicant]
Shi et al., “mvn2vec; Preservation and Collaboration in Multi-View Network Embedding”, arXiv:1801.06597v2, Oct. 30, 2018, 10 pages. [cited by applicant]
Sun et al., “Megan: A Generative Adversarial Network for Multi-View Network Embedding”, Twenty-Eighth International Joint Conference on Artificial Intelligence (IJCAI-19), Aug. 10-16, 2019, Macao, China, 7 pages. [cited by applicant]
Tang et al., “LINE: Large-scale Information Network Embedding”, International Word Wide Web Conference, May 18-22, 2015, Florence, Italy, pp. 1067-1077. [cited by applicant]
Tu et al., “Deep Recursive Network Embedding with Regular Equivalence”, ACM SIGKDD Conference on Knowledge Discover and Data Mining. Aug. 19-23, 2018, London, UK, pp. 2357-2366. [cited by applicant]
Wang et al., “Structural Deep Network Embedding”, ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Aug. 13-17, 2016, San Francisco, CA, 10 pages. [cited by applicant]
Wang et al., “A Survey on Learning to Hash”, arXiv:1606.00185v2, Apr. 22, 2017, 21 pages. [cited by applicant]
Wang et al., “GraphGAN: Graph Representation Learning with Generative Adversarial Nets”, Thirty-Second AAAI Conference on Artificial Intelligence, Feb. 2-7, 2018, New Orleans, Louisiana, pp. 2508-2515. [cited by applicant]
Wang et al., “Learning to Hash on Partial Multi-Modal Data”, Twenty-Fourth International Joint Conference on Artificial Intelligence (IJCAI 2015), Jul. 25-31, 2015, Buenos Aires, Argentina, pp. 3904-3910. [cited by applicant]
Wang et al., “Linked Document Embedding for Classification”, Conference on Information and Knowledge Management (CIKM), Oct. 24-28, 2016, Indianapolis, Indiana, 10 pages. [cited by applicant]
Wang et al., “On Deep Multi-View Representation Learning”, 32 [cited by applicant]
White et al., “Convex Multi-view Subspace Learning”, Twenty-sixth Conference on Neural Information Processing Systems, Dec. 3-8, Lake Tahoe, Nevada, 14 pages. [cited by applicant]
Wu et al., “Efficient Attributed Network Embedding via Recursive Randomized Hashing”, Twenty-Seventh International Joint Conference on Artificial Intelligence (IJCAI-18), Jul. 13-19, 2018, Stockholm, Sweden, pp. 2861-28… [cited by applicant]
Zhang et al., “COSINE: Community-Preserving Social Network Embedding from Information Diffusion Cascades”, Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18), Feb. 2-7, 2018, New Orleans, Louisiana, pp. … [cited by applicant]
Zhang et al., “DANE: Domain Adaptive Network Embedding”, Twenty Eighth International Joint Conference on Artificial Intelligence (IJCAI-19), Aug. 10-16, 2019, Macao, China, pp. 4362-4368. [cited by applicant]
Zhang et al., “Diffusion Maps for Textual Network Embedding”, Thirty-Second Conference on Neural Information Processing Systems, Dec. 2-8, 2018, Montreal, Canada, 11 pages. [cited by applicant]
Zhang et al., “Scalable Multiplex Network Embedding”, Twenty-Seventh International Joint Conference on Artificial Intelligence (IJCAI-18), Jul. 13-19, 2018, Stockholm, Sweden, pp. 3082-3088. [cited by applicant]
Zhang et al., “SINE: Scalable Incomplete Network Embedding”, arXiv:1810.06768v1, Oct. 16, 2018, 10 pages. [cited by applicant]