COMPLEXITY AWARE CALL-STEERING STRATEGY IN HETEROGENEOUS HUMAN/MACHINE CALL-CENTER ENVIRONMENTS

• US 20170214799A1
• Filed: 01/25/2016
• Published: 07/27/2017
• Est. Priority Date: 01/25/2016
• Status: Active Grant

• Current Assignee: Conduent Business Services, LLC (Conduent, Inc.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Perez, Julien, Monet, Nicolas
• Granted Patent: US 9,871,927 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 16/353 into predefined classes
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation
• CPC Class Codes: G06Q 10/10 Office automation; Time management
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: H04M 3/493 Interactive information services, e.g. directory enquiries ; Arrangements therefor, e.g. interactive voice response [IVR] systems or voice portals
• CPC Class Codes: H04M 3/4938 comprising a voice browser which renders and interprets, e.g. VoiceXML
• CPC Class Codes: H04M 3/5166 in combination with interactive voice response systems or voice portals, e.g. as front-ends
• CPC Class Codes: H04M 3/5235 Dependent on call type or called number [DNIS]

Abstract

A method for routing calls suited to use in a call center includes receiving a call from a customer, extracting features from an utterance of the call, and, based on the extracted features, predicting a class and a complexity of a dialogue to be conducted between the customer and an agent. With a routing model, a routing strategy is generated for steering the call to one of a plurality of types of agent (such as to a human or a virtual agent), based on the predicted class and complexity of the dialogue and a cost assigned to the type of agent. A first of the plurality of types of agent is assigned a higher cost than a second of the types of agent. The routing strategy is output.

Claims

1. A method for routing calls comprising:

receiving a call from a customer;

extracting features from an utterance of the call;

based on the extracted features, predicting a class and a complexity of a dialogue to be conducted between the customer and an agent;

with a routing model, generating a routing strategy for steering the call to one of a plurality of types of agent, based on the predicted class and complexity of the dialogue and a cost assigned to the type of agent, a first of the plurality of types of agent being assigned a higher cost than a second of the types of agent; and

outputting the routing strategy,

wherein at least one of the extracting features, predicting the class and the complexity, and generating the routing strategy is performed with a processor.

2. The method of claim 1, further comprising learning a dialogue class model based on a corpus of dialogues, each dialogue in the corpus being annotated with a class from a plurality of classes, the prediction of the class of the dialogue being performed with the dialogue class model.

3. The method of claim 1, wherein the class c of the dialogue is computed according to Equation 1.

4. The method of claim 1, further comprising learning a dialogue complexity model based on a corpus of dialogues, each dialogue in the corpus being annotated with a respective complexity, the prediction of the complexity of the dialogue being performed with the dialogue complexity model.

5. The method of claim 4, wherein the complexity is a function of a number of turns in the dialogue.

6. The method of claim 5, wherein the complexity ι

of the dialogue is predicted according to Equation (2);

(l|d)=ω

₀+ω

₁.φ

(d) 

 

(2)

where φ

(d) is a feature vector derived from the features extracted from the utterance d;

ω

is a vector of parameters to learn; and

ω

₀ is a regularization vector.

7. The method of claim 1, wherein the extracted features comprise n-grams.

8. The method of claim 1, wherein the routing model computes an expectation of success of the dialogue for each of the types of agent, based on the predicted class and complexity of the dialogue, and computes the routing strategy as function of the expectation of success and the assigned cost for each type of agent.

9. The method of claim 8, further comprising, after the dialogue has been conducted:

assessing a success of the dialogue;

computing a difference between the expectation of success, computed by the routing model for the type of agent to which the call has been routed, and the assessed success; and

updating the routing model to reduce the difference.

10. The method of claim 9, wherein the assessment of the success of the dialogue is based on at least one of:

a statistical analysis of at least part of the dialogue; and

customer feedback.

11. The method of claim 1, further comprising assigning the call to an agent based on the routing strategy.

12. The method of claim 1, wherein a first of the types of agent is a human agent and a second of the types of agent is a virtual agent.

13. The method of claim 1, wherein the method includes routing multiple calls from customers among human and virtual agents based on a respective routing strategy for each of the calls.

14. A system comprising memory which stores instructions for performing the method of claim 1 and a processor in communication with the memory for executing the instructions.

15. A computer program product comprising a non-transitory recording medium storing instructions, which when executed on a computer, causes the computer to perform the method of claim 1.

16. A system for routing calls comprising:

a feature extraction component which extracts features from an utterance of a customer call;

a call classification component which predicts a class of a dialogue to be conducted between the customer and an agent, based on the extracted features;

a call complexity component which predicts a complexity of the dialogue to be conducted between the customer and the agent, based on the extracted features;

a routing component which generates a routing strategy for steering the call to one of a plurality of types of agent with a learned routing model, the routing strategy being based on the predicted class of the dialogue, the predicted complexity of the dialogue, and a cost assigned to each type of agent, a first of the plurality of types of agent being assigned a higher cost than a second of the types of agent;

an output component which outputs the routing strategy; and

a processor which implements the components.

17. The system of claim 16, further comprising a feedback component which analyzes feedback on the success of a completed dialogue for updating the model.

18. The system of claim 16, further comprising an update component which updates the routing model based on feedback on success of the dialogue.

19. The system of claim 16, further comprising a virtual agent configured for conducting a dialogue with a customer.

20. A method for generating a system for routing calls among human and virtual agents in a call center, based on an estimate of success of a dialogue to be conducted with an agent, comprising:

learning a dialogue class model for predicting a class of a dialogue to be conducted based on an utterance of a call, the learning being based on prior dialogues conducted between a respective customer and an agent that are annotated by class;

learning a dialogue complexity model for predicting a complexity of the dialogue based on the utterance of the call, the learning being based on prior dialogues conducted between a respective customer and an agent that are annotated by complexity;

learning a dialogue routing model which estimates a success of the dialogue to be conducted between a customer and an agent, when the agent is a human agent, and estimates success of a dialogue conducted between a customer and an agent, when the agent is a real agent, the estimates of success each being a respective function of a class predicted by the dialogue class model and a complexity predicted by the dialogue complexity model;

assigning a cost for weighting the virtual agents differently from the virtual agents,

whereby when calls are received from customers, they are routed among the real and virtual agents as a function of the estimates of success and the assigned cost.