US 11,734,322 B2
Enhanced intent matching using keyword-based word mover's distance
Gregory Kenneth Coulombe, Sherwood Park (CA); Roger C. Meike, Redwood City, CA (US); Cynthia Osmon, Sunnyvale, CA (US); Sricharan Kallur Palli Kumar, Mountain View, CA (US); and Pavlo Malynin, Menlo Park, CA (US)
Assigned to INTUIT, INC., Mountain View, CA (US)
Filed by INTUIT INC., Mountain View, CA (US)
Filed on Nov. 18, 2019, as Appl. No. 16/686,876.
Prior Publication US 2021/0149937 A1, May 20, 2021
Int. Cl. G06F 16/33 (2019.01); G06F 16/35 (2019.01); G06F 40/253 (2020.01); G06F 40/216 (2020.01); G06N 3/08 (2023.01)
CPC G06F 16/3334 (2019.01) [G06F 16/35 (2019.01); G06F 40/216 (2020.01); G06F 40/253 (2020.01); G06N 3/08 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method, comprising:
receiving input of text by a user via a user interface;
determining weights for portions of the text based on a plurality of keywords associated with a domain related to the text, wherein the weights are normalized such that a sum of the weights is a given predetermined value;
generating embeddings of words in the text;
determining a similarity measure between the text and known text associated with an intent by:
determining a first word mover's distance between a first embedding of the embeddings and a first known embedding of a first corresponding word in the known text;
determining a second word mover's distance between a second embedding of the embeddings and a second known embedding of a second corresponding word in the known text;
multiplying the first word mover's distance by a first weight of the weights that corresponds to the first embedding to produce a first product that represents a similarity between the first embedding and the first known embedding weighted according to a first significance to the domain of the first embedding relative to other embeddings of the embeddings;
multiplying the second word mover's distance by a second weight of the weights that corresponds to the second embedding to produce a second product that represents a similarity between the second embedding and the second known embedding weighted according to a second significance to the domain of the second embedding relative to other embeddings of the embeddings; and
calculating the similarity measure based on the first product and the second product;
determining that the intent is associated with the text based on the similarity measure between the text and the known text; and
providing content to the user via the user interface based on the intent.