| CPC G06F 16/63 (2019.01) [G06N 20/00 (2019.01)] | 20 Claims |
|
1. An apparatus comprising:
a memory configured to store:
a set of previous requests;
a set of previous responses, each previous response of the set of previous responses corresponding to at least one previous request of the set of previous requests;
a first set of text associated with a new request; and
a machine learning algorithm configured, when executed by a hardware processor to:
determine rewards associated with the set of previous responses, wherein the rewards are indicators provided by a set of agents that a first plurality of previous responses of the set of previous responses received positive feedback;
determine punishments associated with the set of previous responses, wherein the punishments are indicators provided by the set of agents that a second plurality of previous responses of the set of previous responses received negative feedback;
select between automatically generating a reply to the new request and transmitting the new request to a device of a first agent of the set of agents;
generate a set of groupings based at least in part upon the rewards and the punishments, each grouping of the set of groupings comprising a subset of the set of previous requests and associated with at least one common feature, wherein each previous request of the grouping comprises the at least one common feature;
assign the first set of text to a first grouping of the set of groupings, wherein:
the first grouping comprises a first subset of the set of previous requests;
the assignment is based on a determination that the first set of text comprises the at least one common feature associated with the first grouping; and
the first subset of the set of previous requests is associated with a first subset of the set of previous responses, each previous response of the first subset of the set of previous responses corresponding to at least one previous request of the first subset of the set of previous requests and to the first plurality of previous responses; and
in response to assigning the first set of text to the first grouping, generate a set of suggested responses, wherein each suggested response of the set of suggested responses corresponds to at least one previous response of the first subset of the set of previous responses; and
the hardware processor communicatively coupled to the memory, the hardware processor configured to:
determine that a first user device transmitted a first set of voice signals associated with the new request;
convert the first set of voice signals into the first set of text;
determine to automatically generate the reply to the new request based on a probability that the automatically generated reply comprises information sought by the new request, and the probability is greater than a first threshold;
generate the reply to the new request automatically by:
generating, based on the first set of text, a first response to the new request, wherein the first response to the new request comprises first information sought by the new request;
generating, based on the first set of text, a second response to the new request, wherein the second response to the new request comprises second information sought by the new request;
determining that the reply generated automatically is unsuitable to the new request when the first response is inconsistent with the second response based on at least one of:
the first response to the new request does not comprise the second information;
the second response to the new request does not comprise the first information; and
determining a semantic similarity of the first response to the second response by evaluating a similarity score between the first information and the second information is less than a certain threshold value;
in response to determining that the reply generated automatically is unsuitable to the new request, generate a new punishment to be associated with the reply;
transmit the first set of text to the device of the first agent of the set of agents;
apply the machine learning algorithm to the first set of text to generate the set of suggested responses;
transmit the set of suggested responses to the device of the first agent, wherein the device of the first agent is configured to present the first set of text and the set of suggested responses to the first agent on a display of the device; and
in response to transmitting the set of suggested responses:
determine that the device of the first agent transmitted a second set of voice signals generated by the first agent, the second set of voice signals corresponding to a communication from the first agent to the first user in response to the new request;
convert the second set of voice signals into a second set of text;
store the first set of text in the set of previous requests stored in the memory;
store the second set of text in the set of previous responses stored in the memory, along with an indication that the second set of text corresponds to the first set of text;
relate the second set of text with a first suggested response of the set of suggested responses; and
generate a new reward to be associated with the first suggested response and the second set of text.
|