Link Prediction Spatial Temporal Consistency Dynamic Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing approaches fail to unify spatial and temporal consistency in dynamic network link prediction, leading to suboptimal performance in predicting future network structures.
Innovation Solution
A time-dependent matrix factorization technique is employed to learn feature vectors of nodes by concurrently optimizing temporal fitting and network propagation constraints, ensuring spatial and temporal consistency, and reducing computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing approaches are used for temporal link prediction, then prediction task can be executed, but spatial and temporal consistency cannot be unified leading to suboptimal performance
Solution Approach 1:
The patent merges spatial consistency (network propagation constraint) and temporal consistency (temporal fitting constraint) into a unified optimization framework. The objective function simultaneously minimizes both constraints, allowing the model to learn node embeddings that satisfy both spatial locality and temporal evolution patterns, thereby resolving the contradiction between prediction performance and method complexity.
Solution Approach 2:
The patent transforms the link prediction problem into a parameter optimization problem by learning node embedding vectors. The model changes parameters (node embeddings) to simultaneously satisfy spatial and temporal constraints, using gradient descent to optimize the objective function that combines both consistency requirements, thus improving prediction performance without excessive complexity.
2Reliability
If network propagation constraint and temporal fitting constraint are optimized separately, then each constraint can be satisfied, but concurrent optimization achieves better spatial and temporal consistency
Solution Approach 1:
The patent combines the network propagation constraint (spatial consistency) and temporal fitting constraint (temporal consistency) into a single objective function. This unified optimization approach allows the model to learn node embeddings that simultaneously satisfy both constraints, achieving better overall consistency than separate optimization would provide.
Solution Approach 2:
The patent employs continuous optimization through gradient descent to simultaneously update node embeddings for both spatial and temporal constraints. The continuous nature of the optimization allows the model to smoothly balance both consistency requirements, maintaining useful action across both constraint types without discrete interruptions.
3Measurement precision
If traditional methods are used for finding neighbors of each node, then complete neighbor information can be obtained, but time complexity is high making it unsuitable for large-scale datasets
Solution Approach 1:
The patent makes the node embedding learning process self-service by using the learned embeddings themselves to identify neighbors. Instead of computationally expensive traditional neighbor-finding methods, the model uses similarity in the learned embedding space to implicitly identify relevant neighbors, thereby achieving both accuracy and efficiency.
Solution Approach 2:
The patent replaces the mechanical/computational process of explicitly finding and enumerating all neighbors with a similarity-based approach in embedding space. This substitution uses the learned representations to implicitly capture neighbor relationships, dramatically reducing computational complexity while maintaining accuracy for large-scale networks.
Data Source
AI summary
A computer-implemented method executed by at least one processor for performing link prediction with spatial and temporal consistency by employing a time-dependent matrix factorization technique is presented. The method includes developing, at a plurality of timestamps, relational data of a sequence of network structures each including a plurality of nodes and learning, by the at least one processor, a feature vector of each node of the plurality of nodes of the sequence of network structures by concurrently optimizing a temporal fitting constraint and a network propagation constraint. The method further includes determining a network structure at each timestamp, determining evolutionary patterns at each timestamp, and predicting links in a future network structure based on an evolution of the sequence of network structures within a user-defined sliding window by reducing time complexities of finding neighbors of each node of the plurality of nodes of the sequence of network structures.