Method and apparatus for automatic categorization of calls in a call center environment

1. A system for categorizing a call between an agent and a caller, the system comprising:

at least one processor; and

a memory communicably coupled to the at least one processor, the memory comprising computer executable instructions, which, when executed by the at least one processor implement the method comprising:

receiving a call document comprising text of the call between the agent and the caller;

categorizing the call into a single terminal class from at least one terminal class in a tree-structure of call classes, and one parent of the single terminal class using regressive probability analysis of the call document;

splitting the call document to at least two portions, the at least two portions comprising a call header and a call body;

extracting, using rule-based entity extraction, a mandatory entity from the call header and an optional entity from the call body; and

presenting, on a graphical user interface, the summary comprising the at least one class, the mandatory entity and the optional entity.

2. The system of claim 1 , wherein the regressive probability analysis comprises:

calculating a probability of the call document with respect to the at least one terminal class in a tree-structure of call classes;

identifying, from the at least one terminal class, the single terminal class with the highest probability score as a winner;

calculating the probability of a parent of the winning terminal class, and of other classes having the same ancestor as that of the parent;

identifying the parent of the winning terminal class as the winner among the parent and the other classes having the same ancestor as that of the parent; and

determining the call category as the winning parent and the winning terminal class.

3. The system of claim 1 , wherein the method further comprises verifying the at least one class using progressive probability analysis of the call document,

wherein the progressive probability analysis comprises:

calculating a probability of the call document with respect to at least one top-most class in a tree-structure of call classes;

identifying, from the at least one top-most class, a top-most class with the highest probability score as a winner;

calculating a probability of the call document with respect to at least one child class of the winning top-most class;

identifying, from the at least one child class, a child class with the highest probability score as the winner;

determining the call category as the winning top-most class and the winning child class.

4. The system of claim 1 , wherein the rule-based entity extraction comprises identifying text from the call document as relevant for extracting an entity based on an anticipated phrases in the context of the conversation.

5. The system of claim 4 , wherein the at least two portions comprises a call tail, and wherein the method comprises excluding the call tail from the extracting.

6. A method for categorizing a call between an agent and a caller, the method comprising:

receiving, on a call categorization system, a call document comprising text of the call between the agent and the caller;

categorizing the call into a single terminal class from at least one terminal class in a tree-structure of call classes, and one parent of the single terminal class using regressive probability analysis of the call document;

splitting the call document to at least two portions, the at least two portions comprising a call header and a call body;

extracting, using rule-based entity extraction, a mandatory entity from the call header and an optional entity from the call body; and

presenting, on a graphical user interface, the summary comprising the at least one class, the mandatory entity and the optional entity.

7. The method of claim 6 , wherein the regressive probability analysis comprises:

calculating a probability of the call document with respect to the at least one terminal class;

identifying, from the at least one terminal class, the single terminal class with the highest probability score as a winner;

calculating the probability of a parent of the winning terminal class, and of other classes having the same ancestor as that of the parent;

identifying the parent of the winning terminal class as the winner among the parent and the other classes having the same ancestor as that of the parent; and

determining the call category as the winning parent and the winning terminal class.

8. The method of claim 6 , further comprising verifying the at least one class using progressive probability analysis of the call document,

wherein the progressive probability analysis comprises:

calculating a probability of the call document with respect to at least one top-most class in a tree-structure of call classes;

identifying, from the at least one top-most class, a top-most class with the highest probability score as a winner;

calculating a probability of the call document with respect to at least one child class of the winning top-most class;

identifying, from the at least one child class, a child class with the highest probability score as the winner;

determining the call category as the winning top-most class and the winning child class.

9. The method of claim 6 , wherein the rule-based entity extraction comprises:

identifying text from the call document as relevant for extracting an entity based on an anticipated phrases in the context of the conversation.

10. The method of claim 9 , wherein the at least two portions comprises a call tail, and wherein the method comprises excluding the call tail from the extracting.

11. A non-transitory computer readable medium comprising computer readable instructions that, when executed by a computer, cause the computer to:

receive, on a call categorization system, a call document comprising text of the call between the agent and the caller;

categorize the call into a single terminal class from at least one terminal class in a tree-structure of call classes, and one parent of the single terminal class using regressive probability analysis of the call document;

split the call document to at least two portions, the at least two portions comprising a call header and a call body;

extract, using rule-based entity extraction, a mandatory entity from the call header and an optional entity from the call body; and

present, on a graphical user interface, the summary comprising the at least one class, the mandatory entity and the optional entity.

12. The non-transitory computer readable medium of claim 11 , wherein the regressive probability analysis comprises:

calculating a probability of the call document with respect to the at least one terminal class;

identifying, from the at least one terminal class, the single terminal class with the highest probability score as a winner;

calculating the probability of a parent of the winning terminal class, and of other classes having the same ancestor as that of the parent;

identifying the parent of the winning terminal class as the winner among the parent and the other classes having the same ancestor as that of the parent; and

determining the call category as the winning parent and the winning terminal class.

13. The non-transitory computer readable medium of claim 11 , further comprising verifying the at least one class using progressive probability analysis of the call document,

wherein the progressive probability analysis comprises:

calculating a probability of the call document with respect to at least one top-most class in a tree-structure of call classes;

identifying, from the at least one top-most class, a top-most class with the highest probability score as a winner;

calculating a probability of the call document with respect to at least one child class of the winning top-most class;

identifying, from the at least one child class, a child class with the highest probability score as the winner;

determining the call category as the winning top-most class and the winning child class.

14. The non-transitory computer readable medium of claim 11 , wherein the rule-based entity extraction comprises:

identifying text from the call document as relevant for extracting an entity based on an anticipated phrases in the context of the conversation.

15. The non-transitory computer readable medium of claim 14 , wherein the at least two portions comprises a call tail, and wherein the method comprises excluding the call tail from the extracting.