US 11,941,358 B2
Identifying messages entered into an incorrect conversation
Ethan Russell Elenberg, Jersey City, NJ (US); Cosima Travis, Brooklyn, NY (US); Michael Griffiths, Brooklyn, NY (US); and Kilian Quirin Weinberger, Ithaca, NY (US)
Assigned to ASAPP, INC., New York, NY (US)
Filed by ASAPP, INC., New York, NY (US)
Filed on Jun. 14, 2021, as Appl. No. 17/347,270.
Prior Publication US 2022/0398380 A1, Dec. 15, 2022
Int. Cl. G06F 40/284 (2020.01); G06F 40/30 (2020.01); G06N 3/045 (2023.01); H04L 51/046 (2022.01); H04L 51/21 (2022.01)
CPC G06F 40/284 (2020.01) [G06N 3/045 (2023.01); H04L 51/046 (2013.01)] 20 Claims
OG exemplary drawing
 
8. A computer-implemented method, comprising:
receiving one or more first conversation messages in a first conversation between a user and a first person;
computing a first conversation encoding vector by processing the one or more first conversation messages with a neural network;
receiving one or more second conversation messages in a second conversation between the user and a second person;
computing a second conversation encoding vector by processing the one or more second conversation messages with the neural network;
receiving an entered message entered by the user in the first conversation;
computing a message encoding vector by processing the entered message;
computing a match score indicating a match between the entered message and the first conversation by processing at least one of the message encoding vector, the first conversation encoding vector, or the second conversation encoding vector;
computing a cost score, wherein the cost score indicates a harm from entering the entered message in an incorrect conversation; and
determining, using the match score and the cost score, to warn the user that the entered message may have been entered in an incorrect conversation.