| CPC G06F 16/2246 (2019.01) | 18 Claims |
|
1. A non-transitory computer readable medium comprising instructions which, when executed by one or more hardware processors, causes performance of operations comprising:
storing a first data set as a plurality of node hierarchies in a data repository,
wherein the plurality of node hierarchies comprises a first node hierarchy and a second node hierarchy,
wherein a first node of the first node hierarchy, represents a first entity of a first plurality of entities represented by nodes of the first node hierarchy,
wherein a second node of the first node hierarchy, represents a second entity of the first plurality of entities, and
wherein a position of the first node relative to the second node in the first node hierarchy represents a relationship of the first entity to the second entity;
training a machine learning model to generate rules for modifying nodes of a second node hierarchy, of the plurality of node hierarchies, based at least in part on detecting modifications to the nodes of the first node hierarchy, and
wherein the machine learning model is trained using training datasets from historical node hierarchy modification data, each training dataset of the training datasets comprising:
first node modification data specifying a first modification to a first set of one or more nodes in an initial node hierarchy;
first attribute data specifying attributes of the first set of one or more nodes, the first attribute data including properties of entities represented by the first set of one or more nodes;
second attribute data specifying attributes of a second set of one or more nodes in a target node hierarchy, the second attribute data including properties of entities represented by the second set of one or more nodes;
a first rule specifying a second modification, to the second set of one or more nodes in the target node hierarchy, triggered at least by the first modification, wherein training the machine learning model comprises:
generating, by the machine learning model based on the training data, a prediction representing a predicted rule for modifying the second set of one or more nodes based on the first modification to the first set of one or more nodes; and
based on the prediction generated by the machine learning model, modifying parameters of the machine learning model that determine predictions representing the rules for modifying the second set of one or more nodes based on the first modification to the first set of one or more nodes,
wherein modifying the parameters of the machine learning model comprises backpropagating parameter modifications from an end layer of the machine learning model toward an initial layer of the machine learning model, via intermediate layers of the machine learning model, and
wherein backpropagating the parameter modifications comprises:
modifying first parameter values of the end layer based on a difference between the predicted rule and a particular rule specified in the set of training data; and
modifying second parameter values of a preceding intermediate stage, that is adjacent to the end layer, based on the modification of the first parameter values of the end layer; and
applying the trained machine learning model to a second set of modification data specifying a third modification to a third set of one or more nodes in the first node hierarchy to determine a second rule specifying a fourth modification, to a fourth set of one or more nodes in the second node hierarchy, triggered at least by the third modification.
|