US 12,072,918 B1
Machine learning using knowledge graphs
David Newman, Walnut Creek, CA (US); Omar B. Khan, Richmond, VA (US); and Alexander Joseph Kalinowski, Philadelphia, PA (US)
Assigned to Wells Fargo Bank, N.A., San Francisco, CA (US)
Filed by Wells Fargo Bank, N.A., San Francisco, CA (US)
Filed on Dec. 28, 2021, as Appl. No. 17/646,219.
Int. Cl. G06N 5/02 (2023.01); G06F 16/25 (2019.01); G06F 16/33 (2019.01); G06F 16/338 (2019.01)
CPC G06F 16/3347 (2019.01) [G06F 16/258 (2019.01); G06F 16/338 (2019.01); G06N 5/02 (2013.01)] 17 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
retrieving, over a network device, data from a data source, the data formatted according to a relational database schema;
converting, using at least one processor, the data from the data source into a knowledge graph;
generating a corrupt triple, the corrupt triple not existing in the knowledge graph;
training, using the at least one processor, a neural network using triples from the knowledge graph, wherein a loss function of the neural network is based on a distance between a true triple existing in the knowledge graph before the training and the corrupt triple; and
after the training, storing weights of a hidden layer of the neural network as a vector space representation of entities in the knowledge graph.
 
7. A system comprising:
at least one processor; and
a storage device comprising instructions, which when executed by the at least one processor, configure the at least one processor to perform operations comprising:
retrieving, over a network device, data from a data source, the data formatted according to a relational database schema;
converting the data from the data source into a knowledge graph;
generating a corrupt triple, the corrupt triple not existing in the knowledge graph;
training a neural network using triples from the knowledge graph, wherein a loss function of the neural network is based on a distance between a true triple existing in the knowledge graph before the training and the corrupt triple; and
after the training, storing weights of a hidden layer of the neural network as a vector space representation of entities in the knowledge graph.
 
13. A non-transitory computer-readable medium comprising instructions, which when executed by at least one processor, configure the at least one processor to perform operations comprising:
retrieving, over a network device, data from a data source, the data formatted according to a relational database schema;
converting the data from the data source into a knowledge graph;
generating a corrupt triple, the corrupt triple not existing in the knowledge graph;
training a neural network using triples from the knowledge graph, wherein a loss function of the neural network is based on a distance between a true triple existing in the knowledge graph before the training and the corrupt triple; and
after the training, storing weights of a hidden layer of the neural network as a vector space representation of entities in the knowledge graph.