Game-based link prediction method and system in commutative graphs

Abstract

The invention discloses a game-based link prediction method and system in an exchangeable graph. The method adopts steps: (1) Obtain a data set, and represent the elements in the data set through a graph. The graph is composed of a node set and an edge set; The nodes represent the entities to be predicted, and the edges represent the relationship between the entities to be predicted. The nodes and edges of the data set meet the requirements of the exchangeable graph; (2) use a game-based network evolution model to filter the For edges that do not meet the set requirements, use the filtered graph as the training set; (3) use the training set to iteratively train the probability graph model to obtain a probability graph model with optimal model parameters; (4) use (3) A probabilistic graph model with optimal model parameters performs link prediction. The present invention can be applied in recommendation systems, risk assessment, system planning or social networks. The relationship between entities and entities meets the requirements of exchangeable graphs, and the relationship between entities The relationship prediction is fast and accurate.

Description

technical field [0001] The invention relates to the field of machine learning, in particular to a game-based link prediction method and system in an exchangeable graph. Background technique [0002] In real life, many systems can be modeled by networks whose data structures are graphs. The nodes in the graph represent the entities in the system, and the edges represent the relationships among entities. An exchangeable graph means that in the graph, the order in which edges appear does not affect their distribution, that is, the order of edges is exchangeable. Link prediction (link prediction) refers to the use of known point and edge structures in the network to predict unknown edges, that is, to use the known relationship between entities to predict whether there may be a certain relationship between any two entities. . Link prediction is widely used in recommender systems, risk assessment, and system planning. In commutative graphs, the link prediction problem has no s...

AI Technical Summary

Problems solved by technology

However, since most of these networks are very sparse, there will be a large number of non-existent edges (ie negative samples), which will affect the efficiency and accuracy of model training.

At the same time, some negative sample edges are sometimes not non-existent edges, but may not be observed due to some reasons, such as being missed during observation, etc. These negative samples may immediately establish edges and convert them into positive samples at the next moment , so it is not appropriate to use these as negative samples

On the other hand, in commutative graphs, inference using latent Gaussian processes is generally more complex, less efficient to train, and less parallelizable

Images