CPC G06F 18/217 (2023.01) [G06F 18/214 (2023.01); G06N 20/00 (2019.01); G06T 7/0012 (2013.01); G06T 2207/10116 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06V 2201/03 (2022.01)] | 20 Claims |
1. A method, in a data processing system, for modifying an output of a trained machine learning (ML) computer model based on label co-occurrence statistics to provide an improved ML computer model output, the method comprising:
generating, for each source knowledge data structure in a corpus comprising a plurality of source knowledge data structures, a label vector representation of the source knowledge data structure to thereby generate a plurality of label vector representations;
determining co-occurrence scores for each pairing of labels in a plurality of labels, by generating statistical measures of the co-occurrence of labels in the pairings of labels across the plurality of label vector representations, to thereby generate a label co-occurrence data structure;
receiving an output of the ML computer model, wherein the output is a vector output specifying probability values associated with labels in the plurality of labels;
configuring a knowledge driven reasoning (KDR) computer model with at least one threshold and at least one delta value, wherein the at least one threshold specifies a condition of a co-occurrence of a first label in the output of the ML computer model with a second label in the plurality of labels which, if present, causes the at least one delta value to be applied to modify a probability value associated with the second label in the output of the ML computer model;
executing the KDR computer model on the output of the ML computer model to modify one or more probability values in the output of the ML computer model and generate a modified output of the ML computer model; and
outputting the modified output to a downstream computing system.
|