Graph Data Processing with Initiation and Reception Embedding Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing graph embedding techniques struggle to fully express the data relationships in oriented graphs, leading to information loss during the conversion of graph data into a target graph.
Innovation Solution
A graph data processing method that involves acquiring a target graph with oriented edges representing traffic relationships between objects, and performing multiple iterations to obtain initiation and reception embedding features for each node, which are used to represent the features of initiators and receivers respectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If traditional graph embedding algorithms (DeepWalk, Node2Vec, LINE) are used to map nodes to encoding vectors, then the graph data can be converted to a target graph structure, but information loss occurs because the algorithms cannot fully express the directional traffic relationships between nodes
Solution Approach 1:
The patent segments the encoding vector for each node into two distinct parts: an initiation embedding feature representing the initiator's characteristics and a reception embedding feature representing the receiver's characteristics. This segmentation allows the algorithm to preserve directional information by separately encoding the source and target roles in oriented edges, thereby reducing information loss while maintaining manageable algorithmic complexity through structured decomposition.
Solution Approach 2:
The patent transforms the traditional single encoding vector representation into a two-dimensional feature space by introducing both initiation and reception embedding features for each node. This dimensional expansion enables the model to capture directional relationships in oriented graphs, as each node now has distinct feature representations for its role as initiator versus receiver, thus preserving traffic relationship information that would otherwise be lost.
2Reliability
If oriented edges are introduced to represent traffic relationships with initiator and receiver roles, then information integrity is improved, but the complexity of determining neighborhood nodes and performing iterations increases
Solution Approach 1:
The patent applies local quality by differentiating the treatment of neighborhood nodes based on their directional relationship to the target node. First neighborhood nodes are identified as initiators pointing to the target, while second neighborhood nodes are identified as receivers pointed to by the target. This localized differentiation in how neighborhood nodes are processed and weighted allows the algorithm to maintain high information integrity through accurate directional representation while managing complexity through focused, role-specific processing.
Solution Approach 2:
The patent introduces dynamic iterative processing where the embedding features are updated through multiple iterations, with each iteration refining the initiation and reception features based on the oriented edge relationships. This dynamic approach allows the algorithm to progressively improve information integrity by repeatedly adjusting the feature representations to better capture the directional traffic relationships, while the iterative nature manages complexity through gradual convergence rather than requiring complex one-pass processing.
Data Source
AI summary
A graph data processing method includes: based on object data, acquiring a target graph having M nodes for representing M objects, M being a positive integer greater than 1, and oriented edges for representing presence of traffic relationships between objects corresponding to connected nodes, wherein pointing nodes of the oriented edges are initiators of the traffic relationships, and pointed nodes of the oriented edges are receivers of the traffic relationships; and based on the M nodes and oriented edges between the M nodes, performing multiple iterations to obtain initiation embedding features and reception embedding features corresponding to the M nodes, the initiation embedding features being used for representing features of the initiators, and the reception embedding features being used for representing features of the receivers.