Machine learning model training
Systems and methods for efficiently training a machine learning model are presented. More particularly, using information regarding the relevant neighborhoods of target nodes within a body of training data, the training data can be organized such that the initial state of the training data is relatively easy for a machine learning model to differentiate. Once trained on the initial training data, the training data is then updated such that differentiating between a matching and a non-matching node is more difficult. Indeed, by iteratively updating the difficulty of the training data and then training the machine learning model on the updated training data, the speed that the machine learning model reaches a desired level of accuracy is significantly improved, resulting in reduced time and effort in training the machine learning model.
1. A computer-implemented method for training a machine learning model to identify a matching node for a target node in a corpus graph, the method comprising:
initializing a machine learning model;
accessing a body of training data for training the machine learning model, the body of training data including a plurality of training records, wherein each of the plurality of training records includes a training target node, a corresponding training matching node, and a corresponding training non-matching node;
training the machine learning model using the body of training data until a threshold level of accuracy is achieved with regard to correctly identifying the corresponding training matching nodes for each of the training target nodes of the body of training data;
updating the body of training data to generate an updated body of training data, so as to increase a difficulty of identifying corresponding training matching nodes for the training target nodes included in the updated body of training data;
training the machine learning model using the updated body of training data until the threshold level of accuracy is achieved with regard to correctly identifying the corresponding training matching nodes for each of the training target nodes of the updated body of training data; and
generating an executable machine learning model from the trained machine learning model.
2. The computer-implemented method of claim 1 , wherein the steps of updating the body of training data and training the machine learning model using the updated body of training data is performed for a predetermined number of iterations.
3. The computer-implemented method of claim 2 , wherein the predetermined number of iterations is at least two.
4. The computer-implemented method of claim 2 , wherein:
the body of training data further comprises at least one incomplete training record, the at least one incomplete training record missing at least one of the corresponding training matching nodes or the corresponding training non-matching node; and
the method further comprises completing the body of training data to include the missing node.
5. The computer-implemented method of claim 4 , wherein:
the incomplete training record is missing the corresponding training matching node; and
completing the body of training data comprises:
for each of the at least one incomplete training record of the body of training data:
accessing a relevant neighborhood of the training target node of the incomplete training record; and
selecting a most relevant neighbor node of the relevant neighborhood as the corresponding training matching node of the incomplete training record.
6. The computer-implemented method of claim 5 , wherein:
the incomplete training record is missing the corresponding training non-matching node; and
completing the body of training data further comprises:
for each of the at least one incomplete training record of the body of training data:
randomly selecting a random target node from the corpus graph as the corresponding training non-matching node of the incomplete training record.
7. The computer-implemented method of claim 4 , wherein updating the body of training data comprises:
for each training record of the body of training data:
accessing a relevant neighborhood of the training target node of the training record; and
selecting a neighbor node of the relevant neighborhood as the corresponding training matching node of the training record that is less relevant to the training target node than a previous corresponding training matching node.
8. The computer-implemented method of claim 7 , further comprising:
for each training record of the body of training data:
selecting a second neighbor node of the relevant neighborhood as the corresponding training non-matching node of the training record that is more relevant to the training target node than a previous corresponding training non-matching node.
9. The computer-implemented method of claim 4 , wherein updating the body of training data comprises:
for each training record of the body of training data:
accessing a relevant neighborhood of the training target node of the training record; and
selecting a neighbor node of the relevant neighborhood as the corresponding training non-matching node of the training record that is more relevant to the training target node than a previous corresponding training non-matching node.
10. A non-transitory computer-readable medium bearing computer-executable instructions which, when executed on a computing system comprising at least a processor retrieving instructions from the medium, carry out a method for training a machine learning model to identify a matching node for a target node in a corpus graph, the method comprising:
initializing a machine learning model;
accessing a body of training data for training the machine learning model, the body of training data including a plurality of training records, wherein each of the plurality of training records includes a training target node, a corresponding training matching node, and a corresponding training non-matching node;
training the machine learning model using the body of training data until a threshold level of accuracy is achieved with regard to correctly identifying the corresponding training matching node for each of the training target nodes of the body of training data;
for a predetermined number of iterations:
updating the body of training data to generate an updated body of training data, so as to increase a difficulty of identifying corresponding training matching nodes for the training target nodes included in the updated body of training data; and
training the machine learning model using the updated body of training data until the threshold level of accuracy is achieved with regard to correctly identifying the corresponding training matching node for each of the training target nodes of the updated body of training data; and
generating an executable machine learning model from the trained machine learning model.
11. The non-transitory computer-readable medium of claim 10 , wherein:
the body of training data further comprises at least one incomplete training record, the at least one incomplete training record missing at least one of the corresponding training matching node or the corresponding training non-matching node; and
the method further comprises completing the body of training data to include the missing node.
12. The non-transitory computer-readable medium of claim 11 , wherein:
the at least one incomplete training record is missing the corresponding training matching node; and
completing the body of training data comprises:
for each of the at least one incomplete training record of the body of training data:
accessing a relevant neighborhood of the target node of the incomplete training record; and
selecting a most relevant neighbor node of the relevant neighborhood as the missing corresponding training matching node of the incomplete training record.
13. The non-transitory computer-readable medium of claim 12 , wherein:
the at least one incomplete training record is missing the corresponding training non-matching node; and
completing the body of training data further comprises:
for each of the at least one incomplete training record of the body of training data:
randomly selecting a random target node from the corpus graph as the corresponding training non-matching node of the incomplete training record.
14. The non-transitory computer-readable medium of claim 10 , wherein updating the body of training data comprises:
for each training record of the body of training data:
accessing a relevant neighborhood of the training target node of the training record; and
selecting a neighbor node of the relevant neighborhood as the corresponding matching node of the training record that is less relevant to the training target node than a previous corresponding training matching node.
15. The non-transitory computer-readable medium of claim 10 , wherein updating the body of training data comprises:
for each training record of the body of training data:
accessing a relevant neighborhood of the training target node of the training record; and
selecting a neighbor node of the relevant neighborhood as the corresponding training non-matching node of the training record that is more relevant to the training target node than a previous corresponding training non-matching node.
16. The non-transitory computer-readable medium of claim 10 , wherein the predetermined number of iterations is at least two.
17. A computer system configured to train a machine learning model to identify relevancy between nodes in a corpus graph, the computer system comprising a processor and a memory, wherein the processor executes instructions stored in the memory as part of or in conjunction with additional components to identify relevancy between the nodes in the corpus graph, the additional components comprising:
a data store storing a corpus graph of items;
an executable relevant neighborhood identifier component that, in execution on the computer system, identifies a relevant neighborhood for a target node; and
an executable model trainer component that, in execution on the computer system:
initializes a machine learning model;
accesses a body of training data for training the machine learning model, the body of training data including a plurality of training records, wherein each of the plurality of training records includes a training target node, a corresponding training matching node, and a corresponding training non-matching node;
trains the machine learning model using the body of training data until a threshold level of accuracy is achieved with regard to correctly identifying the corresponding training matching nodes for each of the training target nodes of the body of training data;
for a predetermined number of iterations:
updates the body of training data to generate an updated body of training data so as to increase a difficulty of identifying corresponding training matching nodes for the training target nodes included in the updated body of training data; and
trains the machine learning model using the updated body of training data until the threshold level of accuracy is achieved with regard to correctly identifying the corresponding training matching node for each of the training target nodes of the updated body of training data; and
generates an executable machine learning model from the trained machine learning model.
18. The computer system of claim 17 , wherein the computer system, in updating the body of training data, further:
for each training record of the body of training data:
accesses a relevant neighborhood of the training target node of the training record; and
selects a neighbor node of the relevant neighborhood as the corresponding training matching node of the training record that is less relevant to the training target node than a previous corresponding training matching node.
19. The computer system of claim 18 , wherein the computer system further:
for each training record of the body of training data:
selects a second neighbor node of the relevant neighborhood as the corresponding training non-matching node of the training record that is more relevant to the training target node than a previous corresponding training non-matching node.