System and method for optimizing call flows of a spoken dialog system
View Patent ↗A dialog manager for a spoken dialog system. A decision module selects a path from a plurality of alternative paths for a given call, wherein each path implements one of a plurality of strategies for a call flow. A weighting module weights the path selection decision and is connected to a probability estimator for estimating the probability value that a given one of the plurality of paths is the best-performing path.
1. A dialog manager for a spoken dialog system comprising a computer, a processor, and memory, and comprising:
a decision module for selecting a oath from a plurality of alternative paths for a given call, wherein each path implements one of a plurality of strategies for a call flow; and
a weighting module for weighting the path selection decision, wherein the weighing module is operatively connected to a probability estimator for estimating a probability value that a given path of the plurality of paths is the hest-performing path;
wherein the system is configured to direct incoming calls to each of the alternative paths selected by the decision module, and
wherein the probability estimator includes a reward function R, and wherein the reward function R comprises the expression
R
=
A
-
T
T
A
Wherein A is an automation rate, T is an average handling time, and T A is a trade-off rate for automation rate and average handling time.
2. The dialog manager of claim 1 , wherein the probability estimator includes a parameterized model of the probability density function of the reward function.
3. The dialog manager of claim 1 , wherein the probability estimator includes a dependent probability parameter.
4. The dialog manager of claim 3 , wherein the probability estimator is configured to self-adjust based on the dependent probability parameter.
5. The dialog manager of claim 1 , wherein the decision module selects a path based on a dependent probability parameter.
6. The dialog manager of claim 1 , further comprising:
a dynamic adjustment element configured to dynamically adjust the weights of the weighting module based on data indicating how likely each of the paths will be the best-performing path.
7. The dialog manager of claim 1 , wherein the probability estimator is configured to the winner among two or more paths.
8. The dialog manager of claim 7 , wherein the dialog manager is configured to distribute the calls among the pluralities of strategies in accord with the probability percentages estimated by the probability estimator, the dialog manager comprising:
the decision module being configured to select the path from the plurality of alternative paths for each call, wherein the weighing module is configured to estimate the probability percentage for each path of the plurality of paths, and
wherein the dialog manager is configured to direct each incoming call to a selected path such that the plurality of calls is distributed among alternative paths in accord with the probability percentages selected by the decision module.
9. The system of claim 1 , wherein the decision module includes:
a random generator for selecting a strategy from the plurality of strategies with the weighting of the weighting module.
10. The system of claim 1 wherein the probability estimator is remotely located from the weighting module.
11. A call flow method for a spoken dialog system comprising:
selecting a subset of incoming calls for live implementation of alternate strategies for an activity of a call flow for a spoken dialog system;
configuring a set of weights to determine how the calls will be distributed among the strategies, and
distributing the calls among the strategies in accord with the weights,
wherein the set of weights are distributed in accord with a probability value calculated using a reward function R, and wherein the reward function R comprises the expression
R
=
A
-
T
T
A
Wherein A is an automation rate, T is an average handling time, and T A is a trade-off rate for automation rate and average handling time.
12. The method of claim 11 wherein the method further comprises:
distributing calls among the strategies by winning probabilities.
13. The method of claim 11 wherein the method further comprises:
dynamically adjusting the weights for distributing the calls based on actual performance of the strategies.
14. The method of claim 11 wherein the method further comprises:
dynamically adjusting the weights for distributing the calls based on call parameter.
15. A call flow method for a spoken dialog system comprising a computer, a processor, and memory, comprising:
configuring a plurality of competitor strategies for an activity of a call flow for a spoken dialog system;
implementing the plurality of competitors for the activity in the spoken dialog system; and
wherein the system is configured to select a strategy based on a reward function R, and wherein the reward function R comprises the expression
R
=
A
-
T
T
A
Wherein A is an automation rate, T is an average handling time, and T A is a trade-off rate for automation rate and average handling time.
16. The method of claim 15 , wherein the system is configured perform the method further comprising:
selecting a path from a plurality of alternative paths for a given call, wherein each path implements one of the plurality of competing strategies for the call flow.
17. The method of claim 16 , wherein the system is configured perform the method further comprising:
weighting the path selection decision with weights including a probability value that a given path of the plurality of paths is the best-performing path; and
directing calls to each of the alternative paths in accord with the weights.
18. The method of claim 17 , wherein the probability value is calculated as from the reward function.
19. The method of claim 18 , wherein the probability value is calculated from a parameterized model of a probability density function of the reward function.
20. The method of claim 18 , wherein the system is configured perform the method further comprising:
calculating the probability value from a dependent probability parameter.
21. The method of claim 20 , wherein the wherein the system is configured perform the method further comprising:
automatically self-adjusting the probability value based on the dependent probability parameter.
22. The method of claim 20 , wherein the system is configured perform the method further comprising:
selecting a strategy based on a dependent probability parameter.
23. The method of claim 17 , wherein the system is configured perform the method further comprising:
dynamically adjusting the weights based on data indicating how likely each of the paths will be the best-performing path.
24. The method of claim 16 wherein the system is configured perform the method further comprising:
estimating a winner among two or more paths.
25. The method of claim 16 , wherein the system is configured perform the method further comprising:
distributing the calls among the pluralities of strategies in accord with the probability values;
selecting the path from the plurality of alternative paths for each call, wherein the weights are estimated from a probability percentage for each path of the plurality of paths, and
directing each incoming call to a selected path such that the plurality of calls is distributed among alternative paths in accord with the probability percentages.