US 12,456,454 B2
System and method for a personalized dialogue system using knowledge-based entity services
Hyeongsik Kim, San Jose, CA (US); and Amir Hossein Yazdavar, Manhattan, KS (US)
Assigned to Robert Bosch GmbH, (DE)
Filed by Robert Bosch GmbH, Stuttgart (DE)
Filed on Apr. 28, 2021, as Appl. No. 17/242,411.
Prior Publication US 2022/0351716 A1, Nov. 3, 2022
Int. Cl. G06F 17/00 (2019.01); G06N 3/08 (2023.01); G06N 5/02 (2023.01); G10L 15/02 (2006.01); G10L 25/30 (2013.01)
CPC G10L 15/02 (2013.01) [G06N 3/08 (2013.01); G06N 5/02 (2013.01); G10L 25/30 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method executed by a controller for speech recognition in a system comprising:
parsing a plurality of candidate speech recognition results from a speech input from a user, wherein the parsing is conducted utilizing information associated with a personal knowledge graph corresponding to a user profile of the user and a separate domain-specific knowledge graph that includes a curated collection of background information;
extracting and disambiguating one or more entities from the plurality of candidate speech recognition results, based on natural language understanding (NLU) information utilizing both the personal knowledge graph and the separate domain-specific knowledge graph, wherein the separate domain-specific knowledge graph includes a baseline group and an attention group, wherein the separate domain-specific knowledge graph further includes an attention mechanism configured to, during training, impede overweighting of similar nodes in the baseline group and the attention group, wherein the baseline group integrate embeddings by calculating an average of terms in each dialogue of a subject, wherein disambiguating one or more entities includes linking the one or more entities via utilizing both classifiers and embeddings with a neural network;
selecting and retrieving data indicative of information of the one or more entities from a remote server as one or more relevant entities;
in response to the one or more relevant entities extracted and disambiguated from the speech input, updating the personal knowledge graph associated with the user profile of the user utilizing the data indicative of information of the one or more relevant entities; and
outputting a response to the speech input utilizing the user profile and the NLU information.