US 11,657,819 B2
Selective use of tools for automatically identifying, accessing, and retrieving information responsive to voice requests
Ion Gerald McCusker, Allen, TX (US); Ramakrishna R. Yannam, The Colony, TX (US); Ashwini Patil, Richardson, TX (US); Saahithi Chillara, Allen, TX (US); Ravisha Andar, Plano, TX (US); Emad Noorizadeh, Plano, TX (US); Pravin Kumar Sankari Bhagavathiappan, Chennai (IN); Yogesh Raghuvanshi, Princeton, NJ (US); and Sushil Golani, Charlotte, NC (US)
Assigned to Bank of America Corporation, Charlotte, NC (US)
Filed by BANK OF AMERICA CORPORATION, Charlotte, NC (US)
Filed on Mar. 25, 2021, as Appl. No. 17/212,141.
Prior Publication US 2022/0310086 A1, Sep. 29, 2022
Int. Cl. G09B 7/02 (2006.01); G10L 15/22 (2006.01); G10L 15/10 (2006.01); G06N 20/00 (2019.01); G06F 16/33 (2019.01); G06F 16/2457 (2019.01)
CPC G10L 15/22 (2013.01) [G06N 20/00 (2019.01); G10L 15/10 (2013.01); G06F 16/24578 (2019.01); G06F 16/334 (2019.01); G09B 7/02 (2013.01); G10L 2015/221 (2013.01); G10L 2015/225 (2013.01)] 17 Claims
OG exemplary drawing
 
1. An apparatus comprising:
a memory configured to store a first machine learning algorithm configured, when executed by a hardware processor, to use a set of text associated with a request for information stored within an internal network to select between:
forwarding the request to a device of an agent; and
transmitting a reply to the request, the reply automatically generated based on the set of text and a first source of data stored within the internal network; and
the hardware processor communicatively coupled to the memory, the hardware processor configured to:
receive feedback for a decision made by the first machine learning algorithm, wherein:
the decision made by the first machine learning algorithm comprises at least one of:
a decision to forward the request to the device of the agent; or
a decision to transmit the automatically generated reply to the request; and
the feedback indicates at least one of:
the automatically generated reply comprises the information sought by the request; or
the automatically generated reply does not comprise the information sought by the request;
assign a reward value to the feedback, the reward value comprising at least one of:
a first positive reward value, wherein the decision made by the first machine learning algorithm comprises the decision to forward the request to the device of the agent and the feedback indicates that the automatically generated reply does not comprise the information sought by the request;
a second positive reward value, wherein the decision made by the first machine learning algorithm comprises the decision to transmit the automatically generated reply to the request and the feedback indicates that the automatically generated reply comprises the information sought by the request;
a first negative reward value, wherein the decision made by the first machine learning algorithm comprises the decision to forward the request to the device of the agent and the feedback indicates that the automatically generated reply comprises the information sought by the request; or
a second negative reward value, wherein the decision made by the first machine learning algorithm comprises the decision to transmit the automatically generated reply and the feedback indicates that the automatically generated reply does not comprise the information sought by the request; and
use the reward value to update the first machine learning algorithm; wherein:
the decision made by the first machine learning algorithm comprises the decision to forward the request to the device of the agent;
forwarding the request to the device of the agent comprises:
transmitting the set of text to the device of the agent; and
connecting, over an external network, the device of the agent to a device of the user; and
the hardware processor is further configured to:
monitor for a response to the request transmitted from the device of the agent to the device of the user; and
determine a similarity score between text corresponding to the response to the request transmitted from the device of the agent and text corresponding to the automatically generated reply to the request, wherein the feedback submitted by the agent comprises the similarity score.