US 12,442,659 B1
Systems and methods for supervised machine learning for matching addresses and buildings
Puyang Ma, Bellevue, WA (US)
Assigned to Amazon Technologies, Inc., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Dec. 15, 2022, as Appl. No. 18/066,653.
Int. Cl. G01C 21/00 (2006.01); G01C 21/32 (2006.01); G06Q 10/0835 (2023.01)
CPC G01C 21/3819 (2020.08) [G01C 21/32 (2013.01); G01C 21/3811 (2020.08); G06Q 10/08355 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A method comprising:
receiving, using one or more processors, input data including a first map of a region, a first address associated with the region, and a second address associated with the region, the first map including building outlines representative of buildings included within the region;
receiving, using the one or more processors, first historical data including first indicated delivery locations for first deliveries associated with the first address, and second historical data including second indicated delivery locations for second deliveries associated with the second address;
determining, using the one or more processors and based on the first historical data and the second historical data, a first bounding box including at least some of the first indicated delivery locations and a second bounding box including at least some of the second indicated delivery locations;
determining, using the one or more processors, that a first set of building outlines of the building outlines are at least partially within the first bounding box and a second set of building outlines of the building outlines are at least partially within the second bounding box, wherein a first building outline of the first set of building outlines and a second building outline of the second set of building outlines are both at least partially within both of the first bounding box and the second bounding box;
determining, using the one or more processors, a first set of candidate buildings of the buildings that are associated with the first address and a second set of candidate buildings of the one or more-buildings that are associated with the second address, the first set of candidate buildings including the first set of building outlines, and the second set of candidate buildings including the second set of building outlines;
providing, using the one or more processors, a first feature vector and a second feature vector to a machine learning model, wherein the first feature vector includes features associated with an individual building of the first set of candidate buildings, and wherein the second feature vector includes features shared by all buildings included within the first set of candidate buildings;
providing, using the one or more processors, a third feature vector and a fourth feature vector to the machine learning model, wherein the third feature vector includes features associated with an individual building of the second set of candidate buildings, and wherein the fourth feature vector includes features shared by all buildings included within the second set of candidate buildings;
determining, using the machine learning model, first ranking values associated with the first set of candidate buildings and second ranking values associated with the second set of candidate buildings;
outputting, by the machine learning model and based on the first ranking values and the second ranking values, an indication of a first building of the first set of candidate buildings that is associated with the first address and a second building of the second set of candidate buildings that is associated with the second address; and
generating a second map including an indication of the first address within a first building outline representative of the first building and an indication of the second address within the second building outline that is representative of the second building.