Neural Network Categorization Accuracy With Categorical Graph Neural Networks

A system for performing categorization of a first input into one or more categories arranged in a pre-established taxonomy, the system comprising:

• a first set of computing devices, in aggregate, implementing a graph neural network comprising one or more graph neural network layers, each graph neural network layer utilizing a corresponding set of weight values;
• a second set of computing devices, in aggregate, implementing a first neural network comprising one or more neural network layers, each neural network layer utilizing a corresponding set of weight values, the weight values corresponding to the neural network layers of the first neural network differing from the weight values corresponding to the graph neural network layers of the graph neural network;
• and
• a third set of computing devices categorizing the first input into a first category by performing an interaction between an output of the first set of computing devices and an output of the second set of computing devices to generate scores associated with the categories, the categorization of the first input into the first category being based on a first generated score, associated with the first category, being either greater than others of the generated scores or greater than a threshold;
• wherein the first set of computing devices is provided, as input to the graph neural network, a graph comprising:
• (1) nodes corresponding to the categories and (2) edges between the nodes in accordance with explicit relationships between the categories in the pre-established taxonomy;
• and
• wherein the second set of computing devices is provided, as input to the first neural network, the first input.

One or more computer-readable storage media comprising computer-executable instructions, which, when executed, cause one or more computing devices to:

• receive, as input to a graph neural network, a graph comprising:
• (1) nodes corresponding to categories and (2) edges between the nodes in accordance with explicit relationships between categories in a pre-established taxonomy;
• receive, as input to a first neural network, a first input that is to be categorized into one or more of the categories;
• perform an interaction between an output of the graph neural network and an output of the first neural network to generate scores associated with the categories;
• and
• categorize the first input into a first category based on a first generated score, associated with a first category, being either greater than others of the generated scores or greater than a threshold.