US 12,394,410 B2
Action topic ontology
Emad Noorizadeh, Plano, TX (US); Rajan Jhaveri, Plano, TX (US); and Jennifer Russell, Dallas, TX (US)
Assigned to Bank of America Corporation, Charlotte, NC (US)
Filed by Bank of America Corporation, Charlotte, NC (US)
Filed on Nov. 23, 2022, as Appl. No. 17/993,038.
Prior Publication US 2024/0169979 A1, May 23, 2024
Int. Cl. G06F 40/30 (2020.01); G06F 40/205 (2020.01); G10L 15/18 (2013.01); G10L 15/22 (2006.01)
CPC G10L 15/1815 (2013.01) [G06F 40/205 (2020.01); G06F 40/30 (2020.01); G10L 15/1822 (2013.01); G10L 15/22 (2013.01)] 14 Claims
OG exemplary drawing
 
1. A method for parsing an utterance using an interactive voice response system (“IVR”), said method comprising:
receiving an utterance at the IVR, said utterance received from a user interacting with the IVR via a communication channel;
tokenizing the utterance into a plurality of tokens;
dynamically extracting, from the utterance, a plurality of semantic roles, said dynamically extracting comprising:
determining two or more tokens, included in the plurality of tokens, are candidates for semantic role classification, each of the two or more tokens being a key component of the utterance;
tagging each key component with a semantic role classifier, said semantic role classifiers being selected from a group of semantic role classifiers including: an action, a topic, a topic modifier, an entity, a question and a negation;
determining that one of the semantic role classifiers tagged to a key component is an action, said action being a verb of the utterance;
determining that one of the semantic role classifiers tagged to a key component is a topic, said topic being a subject of the utterance;
determining a first class, within an action topic ontology structured as a static knowledge graph data structure, that corresponds to the token that is classified as an action;
determining a second class, within the action topic ontology structured as a static knowledge graph data structure, that corresponds to the token that is classified as a topic;
using the action topic ontology structured as a static knowledge graph data structure, the first class and the second class to determine a vector that defines a relationship between the token that is classified as an action and the token that is classified as a topic; and
based on the vector, identifying an intent that corresponds to the utterance;
generating a corresponding response to the intent; and
transmitting the corresponding response to the user via the communication channel.