System and method for automatic quality evaluation of interactions

1. A method for automatically calculating an overall evaluation score of an agent for an interaction between a customer and the agent, the method comprising:

receiving, by a processor, an evaluation form for determining the overall evaluation score of agent performance for the interaction, the evaluation form comprising a plurality of automatic questions and a plurality of manual questions;

automatically extracting, by a processor, a set of features from the interaction between the customer and the agent by detecting one or more words, phrases, or topics in a transcript of the interaction, the set of features comprising a plurality of answers to the plurality of automatic questions without manually generated answers to the plurality of manual questions; and

computing the overall evaluation score of the agent based on the plurality of answers to the plurality of automatic questions computed based on the set of features;

wherein the computing the overall evaluation score based on the set of features comprises: supplying the set of features to a deep neural network trained based on the evaluation form; and computing the overall evaluation score by forward propagating the set of features through the deep neural network;

wherein the set of features from the interaction consist of the plurality of answers to the automatic questions; and

wherein:

the plurality of automatic questions each comprises a question relating to a characteristic of the agent performance for the interaction that is answerable to a sufficient level of accuracy via automatic processes; and

the plurality of manual questions each comprises a question relating to a characteristic of the agent performance for the interaction that is not answerable to a sufficient level of accuracy without being answered manually by a human;

wherein the set of features from the interaction further comprises automatically generated scores for the manual questions; and

wherein each of the automatically generated scores for the manual questions is generated by: identifying a closest question-answer-document triple from a plurality of question-answer-document triples, the closest question-answer-document triple being most similar to a manual question and the interaction; and returning a score corresponding to the answer associated with the closest question-answer-document triple.

2. The method of claim 1 , wherein the set of features from the interaction comprises topics detected in the interaction.

3. The method of claim 1 , further comprising:

receiving a second evaluation form, the second evaluation form comprising a second plurality of automatic questions and a second plurality of manual questions, wherein the interaction is associated with one evaluation form;

creating a super set of automatic questions comprising the plurality of automatic questions of the evaluation form and the second plurality of automatic questions of the second evaluation form;

creating a super set of manual questions comprising the plurality of manual questions of the evaluation form and the second plurality of manual questions of the second evaluation form; and

computing an index vector, wherein each value in the index vector corresponds to a question of the super set of manual questions, wherein each value of the index vector indicates whether the evaluation form associated with the interaction includes the corresponding question of the super set of manual questions,

wherein the set of features further comprise the index vector.

4. The method of claim 1 , wherein each of the automatically generated scores for the manual questions is generated by: detecting one or more topics in the interaction; computing a semantic similarity between a manual question and the detected one or more topics in the interaction; and generating a score in accordance with the semantic similarity.

5. The method of claim 1 , wherein the deep neural network is trained based on answers to the automatic questions of the evaluation form and manually generated answers to the manual questions of the evaluation form.

6. A system comprising:

a processor;

a non-transitory memory storing instructions that, when executed by the processor, cause the processor to:

receive an evaluation form for determining an overall evaluation score of agent performance, the evaluation form comprising a plurality of automatic questions and a plurality of manual questions;

receive an interaction between a customer and an agent for evaluation of the agent performance;

automatically extract a set of features from the interaction by detecting one or more words, phrases, or topics in a transcript of the interaction, the set of features comprising a plurality of answers to the plurality of automatic questions without manually generated answers to the plurality of manual questions; and

compute the overall evaluation score of the agent based on the plurality of answers to the plurality of automatic questions computed based on the set of features;

wherein the memory further stores instructions that, when executed by the processor, cause the processor to compute the overall evaluation score based on the set of features by:

supplying the set of features to a deep neural network trained based on the evaluation form; and

computing the overall evaluation score by forward propagating the set of features through the deep neural network;

wherein the set of features from the interaction consist of the plurality of answers to the automatic questions; and

wherein:

the plurality of automatic questions each comprises a question relating to a characteristic of the agent performance for the interaction that is answerable to a sufficient level of accuracy via automatic processes; and

the plurality of manual questions each comprises a question relating to a characteristic of the agent performance for the interaction that is not answerable to a sufficient level of accuracy without being answered manually by a human,

wherein the set of features from the interaction further comprises automatically generated scores for the manual questions, and

wherein the memory further stores instructions that, when executed by the processor, cause the processor to generate the automatically generated scores for the manual questions by:

identifying a closest question-answer-document triple from a plurality of question-answer-document triples, the closest question-answer-document triple being most similar to a manual question and the interaction; and

returning a score corresponding to the answer associated with the closest question-answer-document triple.

7. The system of claim 6 , wherein the set of features from the interaction comprises topics detected in the interaction.

8. The system of claim 6 , wherein the memory further stores instructions that, when executed by the processor, cause the processor to:

receive a second evaluation form, the second evaluation form comprising a second plurality of automatic questions and a second plurality of manual questions;

create a super set of automatic questions comprising the plurality of automatic questions of the evaluation form and the second plurality of automatic questions of the second evaluation form;

create a super set of manual questions comprising the plurality of manual questions of the evaluation form and the second plurality of manual questions of the second evaluation form; and

compute an index vector, wherein each value in the index vector corresponds to a question of the super set of manual questions, wherein each value of the index vector indicates whether the evaluation form associated with the interaction includes the corresponding question of the super set of manual questions,

wherein the set of features further comprise the index vector.

9. The system of claim 6 , wherein the memory further stores instructions that, when executed by the processor, cause the processor to generate the automatically generated scores for the manual questions by: detecting one or more topics in the interaction; computing a semantic similarity between a manual question and the detected one or more topics in the interaction; and generating a score in accordance with the semantic similarity.

10. The system of claim 6 , wherein the deep neural network is trained based on answers to the automatic questions of the evaluation form and manually generated answers to the manual questions of the evaluation form.