US 12,242,801 B2
Blending graph predictions
Thanh Lam Hoang, Maynooth (IE); Gabriele Picco, Dublin (IE); Yufang Hou, Dublin (IE); Young-Suk Lee, Mahopac, NY (US); Lam Minh Nguyen, Ossining, NY (US); Dzung Tien Phan, Pleasantville, NY (US); Vanessa Lopez Garcia, Dublin (IE); and Ramon Fernandez Astudillo, White Plains, NY (US)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Feb. 8, 2022, as Appl. No. 17/666,827.
Prior Publication US 2023/0252234 A1, Aug. 10, 2023
Int. Cl. G06F 40/205 (2020.01); G06F 40/30 (2020.01)
CPC G06F 40/205 (2020.01) [G06F 40/30 (2020.01)] 18 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
receiving a set of graph predictions corresponding to an input text, where each graph prediction of the set of graph predictions is generated by a different machine learning model and each graph prediction is a semantic representation graph of the input text;
for each graphs prediction:
comparing a current graph prediction to each other graph prediction of the set of graph predictions;
counting, for each other graph prediction, a vote for each common edge and node; and
blending, based on the comparing and the counting, the current graph prediction to generate a first blended graph; and
adding the first blended graphs to a plurality of candidate blended graphs, where nodes and edges of the candidate blended graphs have respective selection metric values, generated using a selection metric function, that meet a minimum threshold; and
selecting as an output blended graph a candidate blended graph of the plurality of candidate blended graphs having a highest total combination of selection metric values among the plurality of candidate blended graphs.