US 12,455,923 B2
Intelligent collaborative decision generation system with iterative prediction of link set in knowledge graph
Christophe Gueret, Donabate (IE); Sumit Pai, Palmerstown (IE); Luca Costabello, Newbridge (IE); and Rory McGrath, Kildare (IE)
Assigned to Accenture Global Solutions Limited, Dublin (IE)
Filed by Accenture Global Solutions Limited, Dublin (IE)
Filed on Jun. 12, 2023, as Appl. No. 18/333,093.
Claims priority of provisional application 63/351,560, filed on Jun. 13, 2022.
Prior Publication US 2023/0401258 A1, Dec. 14, 2023
Int. Cl. G06F 16/901 (2019.01); G06N 5/025 (2023.01)
CPC G06F 16/9024 (2019.01) [G06N 5/025 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A system comprising at least one memory for storing computer instructions and a processing circuitry in communication with the at least one memory, the processing circuitry, when executing the computer instructions, is configured to:
ingest a knowledge graph comprising a plurality of nodes and a plurality of ground-truth edges to generate a current knowledge graph;
initiate a predicted link set associated with the current knowledge graph;
iteratively, until an iteration stoppage condition is met:
convert, by a graph machine learning model, the current knowledge graph into embedding vectors in a multidimensional embedding space, wherein the graph machine learning model includes:
a scoring layer configured to score positive triples and negative triples of the current knowledge graph;
a plug-in layer configured to alter the scores output by the scoring layer based on confidence of edges of a feedback knowledge graph generated based on user feedback on previous predicted link set and explanation on the previous predicted link set, and
a loss layer configured to aim at increasing margin between the scores of the positive triples and the negative triples based on output of the plug-in layer, for optimizing parameters of the graph machine learning model in generating the embedding vectors;
generate a set of single predicted edges, based on the embedding vectors and a query, characterized by normalized linkage scores;
rank the set of single predicted edges based on the normalized linkage scores in a ranking order;
determine whether the predicted link set is empty;
add at least one single predicted edge of the set of single predicted edges according to the ranked order to the predicted link set, based on the predicted link set is empty;
evaluate a criterion for addition of the at least one single predicted edge to the predicted link set, based on the predicted link set is non-empty;
add the at least one single predicted edge to the predicted link set based on the evaluation of the criterion; and
add the at least one single predicted edge of the set of single predicted edges to expand the current knowledge graph; and
after the iteration stoppage condition is met, automatically generate a plain text description of the predicted link set, wherein the plain text description of the predicted link set is generated based on a history of the iterations for expansion of the predicted link set.