Threshold-based adaptive ontology and knowledge graph modification using generative artificial intelligence

A computing system comprising:

• one or more processors;
• one or more non-transitory, computer-readable storage media storing instructions that, when executed by the one or more processors, cause the computing system to:
• retrieve, from a node database, a node map representing (1) a set of nodes and (2) a set of node relationships between one or more nodes of the set of nodes,
• wherein each node of the set of nodes includes a particular object of a set of objects, and
• wherein the node relationships of the set of node relationships include relationship data structures representing relationships between nodes of the set of nodes;
• provide a representation of the node map to a large-language model to determine a first perturbation comprising at least one of (1) a node perturbation in a first node component of a first node of the set of nodes or (2) a relationship perturbation in a first node relationship of the set of node relationships,
• wherein the node perturbation includes a modification in a first object of the first node, and
• wherein the relationship perturbation includes a modification in a first relationship of the first node relationship;
• apply the first perturbation to the node map to generate a test node map,
• wherein the test node map represents at least one of:
• (1) the first node including the node perturbation or (2) the first node relationship including the relationship perturbation;
• retrieve, from a training database, a training input and a ground-truth output;
• generate, according to the node map and using the training input, a first output;
• generate, based on the first output and the ground-truth output, a first loss metric value associated with the node map;
• generate, according to the test node map and using the first perturbation and the training input, a second output;
• generate, based on the second output and the ground-truth output, a second loss metric value associated with the test node map;
• determine a perturbation quality value based on a difference between the first loss metric value and the second loss metric value;
• determine that the perturbation quality value satisfies a threshold quality criterion;
• and
• in response to determining that the perturbation quality value satisfies the threshold quality criterion, update the node database to include the test node map in lieu of the node map.

One or more non-transitory, computer-readable storage media storing instructions that, when executed by one or more processors of a computing system, cause the computing system to:

• obtain a node map representing (1) a set of nodes and (2) a set of node relationships between one or more nodes of the set of nodes, and
• wherein the node relationships of the set of node relationships represent relationships between nodes of the set of nodes;
• provide a representation of the node map to a large-language model to determine a first perturbation of at least one component of the node map;
• apply the first perturbation to the node map to generate a test node map consistent with the first perturbation;
• retrieve, from a training database, a training input and a ground-truth output;
• generate, according to the node map and using the training input, a first output;
• generate, based on the first output and the ground-truth output, a first loss metric value associated with the node map;
• generate, according to the test node map and using the first perturbation and the training input, a second output;
• generate, based on the second output and the ground-truth output, a second loss metric value associated with the test node map;
• determine a perturbation quality value based on a difference between the first loss metric value and the second loss metric value;
• determine that the perturbation quality value satisfies a threshold quality criterion;
• and
• in response to determining that the perturbation quality value satisfies the threshold quality criterion, transmit, to a user device, the test node map in lieu of the node map.