Tensor-based predictions from analysis of time-varying graphs

A method for analyzing time-varying graphs respectively comprising nodes representing elements in a network, edges representing transactions between elements and data associated with the nodes and the edges, the method comprising:

• constructing adjacency matrices describing each node and edge of each time-varying graph for stacking into an adjacency tensor;
• constructing feature matrices describing the data of each time-varying graph for stacking into a feature tensor;
• partitioning the adjacency and feature tensors into adjacency and feature training tensors and into adjacency and feature validation tensors, respectively;
• creating an embedding model for computing an embedding of the adjacency and feature training tensors and a prediction model using the embedding for computing a prediction;
• validating the embedding and prediction models using the adjacency and feature validation tensors to identify an optimized embedding-prediction model pair;
• computing the predictions of qualities of relationships between the individuals at the future time using the optimized embedding-prediction pair;
• and
• preemptively managing computing resources relative to the relationships at the future time in accordance with the predicted qualities.

A computer program product, comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to perform a method of analyzing time-varying graphs respectively comprising nodes representing elements in a network, edges representing transactions between elements and data associated with the nodes and the edges, the method comprising:

• constructing adjacency matrices describing each node and edge of each time-varying graph for stacking into an adjacency tensor;
• constructing feature matrices describing the data of each time-varying graph for stacking into a feature tensor;
• partitioning the adjacency and feature tensors into adjacency and feature training tensors and into adjacency and feature validation tensors, respectively;
• creating an embedding model for computing an embedding of the adjacency and feature training tensors and a prediction model using the embedding for computing a prediction;
• and
• validating the embedding and prediction models using the adjacency and feature validation tensors to identify an optimized embedding-prediction model pair;
• computing the predictions of qualities of relationships between the individuals at the future time using the optimized embedding-prediction pair;
• and
• preemptively managing computing resources relative to the relationships at the future time in accordance with the predicted qualities.

A method for analyzing time-varying graphs respectively comprising nodes representing individuals in a social network, edges representing friendships between individuals and data associated with the nodes and the edges, the method comprising:

• constructing, using a processor, adjacency and feature matrices describing each node and edge of each time-varying graph for stacking into an adjacency tensor and describing the data of each time-varying graph for stacking into a feature tensor, respectively;
• partitioning, using the processor, the adjacency and feature tensors into adjacency and feature training tensors and into adjacency and feature validation tensors, respectively;
• creating, using the processor, an embedding model for computing an embedding of the adjacency and feature training tensors and a prediction model using the embedding for computing predictions of qualities of relationships between the individuals at a future time;
• validating, using the processor, the embedding and prediction models using the adjacency and feature validation tensors to identify an optimized embedding-prediction model pair;
• computing, using the processor, the predictions of qualities of relationships between the individuals at the future time using the optimized embedding-prediction pair;
• and
• preemptively managing, using the processor, computing resources relative to the relationships at the future time in accordance with the predicted qualities.

A method for analyzing time-varying graphs respectively comprising nodes representing entities in a business network, edges representing transactions between entities and data associated with the nodes and the edges, the method comprising:

• constructing, using a processor, adjacency and feature matrices describing each node and edge of each time-varying graph for stacking into an adjacency tensor and describing the data of each time-varying graph for stacking into a feature tensor, respectively;
• partitioning, using the processor, the adjacency and feature tensors into adjacency and feature training tensors and into adjacency and feature validation tensors, respectively;
• creating, using the processor, an embedding model for computing an embedding of the adjacency and feature training tensors and a prediction model using the embedding for computing predictions of qualities of relationships between the individuals at a future time;
• validating, using the processor, the embedding and prediction models using the adjacency and feature validation tensors to identify an optimized embedding-prediction model pair;
• computing, using the processor, the predicted trustworthiness of transactions between entities at the future time using the optimized embedding-prediction pair;
• and
• preemptively authorizing, using the processor, or blocking the transactions at the future time in accordance with the predicted trustworthiness.