US 12,259,916 B1
Enhanced search autocompletion framework
Liwen You, San Jose, CA (US); and V Divya Bhargavi, Atlanta, GA (US)
Assigned to Amazon Technologies, Inc., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Jun. 8, 2023, as Appl. No. 18/331,468.
Int. Cl. G06F 16/33 (2019.01); G06F 16/334 (2025.01); G06F 16/338 (2019.01)
CPC G06F 16/3344 (2019.01) [G06F 16/338 (2019.01)] 20 Claims
OG exemplary drawing
 
4. A method comprising:
receiving, at a search autocompletion system from a search application, prefix data comprising one or more characters representing a prefix;
determining, by the search autocompletion system, autocompletion candidate data representing:
one or more first correspondences between the prefix and one or more nodes in a knowledge graph representing a target search space;
one or more second correspondences between the prefix and historical query data; and
one or more third correspondences between the prefix and user historical interaction data;
executing, by the search autocompletion system and using the autocompletion candidate data as input, a machine-learned autocompletion candidate determination model to:
execute a graph neural network using, as first input, a first subset of the autocompletion candidate data comprising a first subset of the one or more first correspondences, a first subset of the one or more second correspondences, a first subset of the one or more third correspondences, and second input generated by a language model to generate first feature output;
execute the language model using, as second input, a second subset of the autocompletion candidate data comprising a second subset of the one or more first correspondences, a second subset of the one or more second correspondences, a second subset of the one or more third correspondences, and third input generated by the graph neural network to generate second feature output;
determine, based at least in part on the first feature output and the second feature output, scores for individual candidates represented in at least one of the one or more first correspondences, the one or more second correspondences, or the one or more third correspondences; and
determine, based at least in part on the scores for the individual candidates, an ordered set of autocompletion candidates; and
transmitting, by the search autocompletion system to the search application for presentation to a user, the ordered set of autocompletion candidates.