Multi-View Network Embedding for Incomplete Graph Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-view network embedding methods fail to effectively handle incomplete data, leading to suboptimal performance in tasks such as node classification, clustering, and link prediction, as they assume all nodes are present in all views, which is often not the case in real-world applications.
Innovation Solution
A unified network embedding approach that learns embeddings by enforcing data consistency between different views through deep latent subspace learning and preserving data proximity within each view using graph Laplacian, while handling partial multi-view data without requiring complete information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing multi-view network embedding methods are used, then network embedding can be generated, but they fail to handle incomplete data effectively because they assume all nodes are present in all views
Solution Approach 1:
The patent segments the embedding space into view-specific embedding spaces and a common embedding space. Each view maintains its own embedding space for nodes present in that view, while a common embedding space captures shared representations. This segmentation allows the method to handle incomplete data by treating each view independently while still enabling cross-view relationships through the common space.
Solution Approach 2:
The patent introduces a common embedding space as an intermediary that mediates between different view-specific embedding spaces. This common space serves as a bridge that allows information to be shared and transferred between views, enabling the method to handle cases where nodes are missing from certain views while maintaining overall network structure.
2Quantity of substance
If data preprocessing is performed to fill in missing data, then complete data can be obtained, but the performance does not achieve good results
Solution Approach 1:
Instead of attempting to completely fill in all missing data (excessive action), the patent adopts partial action by working directly with the available data in each view. The method processes only the data that is actually present, using the view-specific embedding spaces to accommodate incomplete information without requiring imputation of missing values.
3Stability of the object's composition
If data is removed that suffers from missing information, then complete views can be maintained, but nodes cannot be mapped to embedding vectors
Solution Approach 1:
The patent creates a universal embedding framework that serves multiple functions: it can handle complete views, incomplete views, and generate embeddings for all nodes regardless of which views they appear in. The common embedding space provides a universal representation that works across all views, while view-specific spaces handle view-particular characteristics.
Data Source
AI summary
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.