Optimizing CogBot retraining
The continual retraining of Cognitive Bots (“CogBots”) allows for the adaptation and evolution to ever changing environments. In this retraining process a CogBot response model continually searches the response space for potential responses in which it may shift. An approach for optimizing such retraining of CogBots may be presented herein. The approach may include receiving a prompt at a CogBot retraining framework. The approach may include analyzing the prompt and determining potential responses to the prompt. The approach may include generating a dialogue benchmark for each of the potential responses. The approach may further include generating a decision shift score for the prompt. Further, the approach may additionally include updating the CogBot retraining framework based on the generated decision shift score.
1 . A computer-implemented method for optimizing Cognitive Bot (CogBot) retraining, the method comprising:
establishing a framework for understanding shift behaviors of a CogBot having a retraining shift algorithm for continuous, automated improvement and updating the retraining shift algorithm as needed, the framework including an algorithm analysis module and a retraining optimization module;
identifying, within the framework, a prompt at a CogBot response model;
analyzing, by the algorithm analysis module, the prompt and a current response prepared by the CogBot by:
generating, by the algorithm analysis module, one or more potential decisions for the prompt, based at least in part on analyzing the prompt and the current response; and
calculating, by the retraining optimization module, a dialogue benchmark for each of the one or more potential decisions with respect to the current response;
responsive to a first potential decision for the prompt being above a dialogue benchmark threshold, shifting, by the retraining optimization module, the current response to the first potential decision for the prompt, the first potential decision having a highest dialogue benchmark among the one or more potential decisions;
determining a shift decision score for shifting the current response to the first potential decision; and
updating the retraining shift algorithm of the CogBot response model based at least in part on the shift decision score.
2 . The computer-implemented method of claim 1 , wherein determining a shift decision score further comprises:
generating a wrong decision score for the prompt, based at least in part on the generated one or more potential decisions; and
generating a shift probability for the prompt.
3 . The computer-implemented method of claim 2 , wherein determining a wrong decision score further comprises:
generating an unnecessary shift probability, wherein the unnecessary shift probability is a probability the CogBot shifts to a potential decision with a lower dialogue benchmark.
4 . The computer-implemented method of claim 2 , wherein determining a wrong decision score further comprises:
generating a missing shift probability, wherein the missing shift probability is a probability the CogBot fails to shift to a potential decision resulting in a higher dialogue benchmark.
5 . The computer-implemented method of claim 1 , wherein determining a shift decision score further comprises:
configuring the one or more potential decisions as a Poisson process, wherein the CogBot has one or more instances to shift into a decision with a dialogue benchmark above a calculated threshold.
6 . The computer-implemented method of claim 1 , wherein:
the CogBot is a primary CogBot and the identified prompt and the current response are provided to the primary CogBot by a secondary CogBot stored on a mobile device;
the framework is operational on a host device, separate from the mobile device; and
shifting the current response establishes a shift update, which is provided to the secondary CogBot from the host device via the primary CogBot.
7 . A computer system for optimizing CogBot retraining, the system comprising:
a memory; and
a processor in communication with the memory, the processor being configured to perform operations to:
establish a framework for understanding behaviors of the CogBot having a retraining shift algorithm for continuous, automated improvement and updating the retraining shift algorithm as needed, the framework including an algorithm analysis module and a retraining optimization module;
identify, within the framework, a prompt at a CogBot response model;
analyze, by the algorithm analysis module, the prompt and a current response prepared by the CogBot by:
generate, by the algorithm analysis module, one or more potential decisions for the prompt, based at least in part on analyzing the prompt and the current response; and
calculate, by the retraining optimization module, a dialogue benchmark for each of the one or more potential decisions with respect to the current response;
responsive to a first potential decision for the prompt being above a dialogue benchmark threshold, shift, by the retraining optimization module, the current response to the first potential decision for the prompt, the first potential decision having a highest dialogue benchmark among the one or more potential decisions;
determine a shift decision score for shifting the current response to the first potential decision; and
update the retraining shift algorithm of the CogBot response model based at least in part on the shift decision score.
8 . The computer system of claim 7 , the processor further configured to perform operations to:
generate a wrong decision score for the prompt, based at least in part on the generated one or more potential decisions; and
generate a shift probability for the prompt.
9 . The computer system of claim 8 , wherein determining a wrong decision score further comprises operations to:
generate an unnecessary shift probability, wherein the unnecessary shift probability is a probability the CogBot shifts to a potential decision with a lower dialogue benchmark.
10 . The computer system of claim 8 , wherein determining a wrong decision score further comprises operations to:
generate a missing shift probability, wherein the missing shift probability is a probability the CogBot fails to shift to a potential decision resulting in a higher dialogue benchmark.
11 . The computer system of claim 7 , wherein determining a shift decision score further comprises operations to:
configure the one or more potential decisions as a Poisson process, wherein the CogBot has one or more instances to shift into a decision with a dialogue benchmark above a calculated threshold.
12 . A computer program product comprising one or more non-transitory computer-readable storage media having program instructions embodied therewith which, when executed by a processor causes the processor to perform a method comprising:
establishing a framework for understanding behaviors of a CogBot having a retraining shift algorithm for continuous, automated improvement and updating the retraining shift algorithm as needed, the framework including an algorithm analysis module and a retraining optimization module;
identifying, within the framework, a prompt at a CogBot response model;
analyzing, by the algorithm analysis module, the prompt and a current response prepared by the CogBot by:
generating, by the algorithm analysis module, one or more potential decisions for the prompt, based at least in part on analyzing the prompt and the current response; and
calculating, by the retraining optimization module, a dialogue benchmark for each of the one or more potential decisions with respect to the current response;
responsive to a first potential decision for the prompt being above a dialogue benchmark threshold, shifting, by the retraining optimization module, the current response to the first potential decision for the prompt, the first potential decision having a highest dialogue benchmark among the one or more potential decisions;
determining a shift decision score for shifting the current response to the first potential decision; and
updating the retraining shift algorithm of the CogBot response model based at least in part on the shift decision score.
13 . The computer program product of claim 12 , further causing the processor to perform a method comprising:
generating a wrong decision score for the prompt, based at least in part on the generated one or more potential decisions; and
generating a shift probability for the prompt.
14 . The computer program product of claim 12 , wherein determining a shift decision score further comprises:
generating a missing shift probability, wherein the missing shift probability is a probability the CogBot fails to shift to a potential decision resulting in a higher dialogue benchmark.
15 . The computer program product of claim 12 , wherein determining a wrong decision score further comprises:
configuring the one or more potential decisions as a Poisson process, wherein the CogBot has one or more instances to shift into a decision with a dialogue benchmark above a calculated threshold.