System and method for identification of intent segment(s) in caller-agent conversations
Identification of an intent of a conversation can be useful for real-time or post-processing purposes. According to example embodiments, a method, and corresponding apparatus of identifying at least one intent-bearing utterance in a conversation, comprises determining at least one feature for each utterance among a subset of utterances of the conversation; classifying each utterance among the subset of utterances, using a classifier, as an intent classification or a non-intent classification based at least in part on a subset of the at least one determined feature; and selecting at least one utterance, with intent classification, as an intent-bearing utterance based at least in part on classification results by the classifier. Through identification of an intent bearing utterance, a call center for example, can provide improved service for callers through, for example, more effective directing of a call to a live agent.
1. A method of identifying at least one intent-bearing utterance in a conversation between a human agent, associated with a call center, and a caller, the method comprising:
determining a plurality of features for each utterance among a set of utterances of the conversation between the human agent and the caller, the features including one of an identity associated with the utterance between the human agent and the caller, a position of the utterance within the conversation, and a word located before or after the utterance within the conversation;
classifying each utterance among the set of utterances, using a classifier, as an intent-bearing classification or a non-intent-bearing classification by generating multiple state sequences for each set of utterances, each state sequence including a plurality of states, each of the plurality of states corresponding to a respective utterance among the set of utterances and each of the plurality of states representing a probability that the respective utterance is intent-bearing, each utterance being associated with respective probabilities of the multiple state sequences;
generating a ranking of each utterance based on a combination of the respective probabilities of the multiple state sequences, the ranking being a relative measure that the respective utterance is intent-bearing; and
marking, in a customer relationship management (CRM) system, each utterance as being an intent-bearing utterance representing an intent of the conversation based on the ranking.
2. The method according to claim 1 , wherein selecting at least one utterance includes:
selecting the at least one utterance based on a highest occurrence of intent-bearing classification among the multiple probabilities being above a particular threshold.
3. The method according to claim 1 , wherein the word or expression frequently used prior to, within, or following an intent bearing utterance being provided to the classifier manually through rules or automatically identified in a data driven manner.
4. The method of claim 1 , wherein:
the features include two or more of the following: an identity associated with the utterance between the human agent and the caller, the position of the utterance within the conversation, and the word located before or after the utterance within the conversation; and wherein
classifying each utterance further includes classifying each utterance among the set of utterances, using a classifier, as an intent-bearing classification or a non-intent-bearing classification based on the two or more determined features.
5. An apparatus for identifying at least one intent-bearing utterance in a conversation between a human agent, associated with a call center, and a caller, the apparatus comprising:
a memory with computer code instructions stored thereon; and
a processor,
the memory, with the computer code instructions, and the processor being configured to:
determine a plurality of features for each utterance among a set of utterances of the conversation between the human agent and the caller, the features including one of an identity associated with the utterance between the human agent and the caller, a position of the utterance within the conversation, and the word located before or after the utterance within the conversation;
classify each utterance among the set of utterances, using a classifier, as an intent-bearing classification or a non-intent-bearing classification by generating multiple state sequences for each set of utterances, each state sequence including a plurality of states, each of the plurality of states corresponding to a respective utterance among the set of utterances and each of the plurality of states representing a probability that the respective utterance is intent-bearing, each utterance being associated with respective probabilities of the multiple state sequences;
generate a ranking of each utterance based on a combination of the respective probabilities of the multiple state sequences, the ranking being a relative measure that the respective utterance is intent-bearing; and
select an utterance as being an intent-bearing utterance representing an intent of the conversation based on the ranking.
6. The apparatus according to claim 5 , wherein in selecting at least one utterance the memory with the computer code instructions and the processor are configured to:
select the at least one utterance based on a highest occurrence of intent-bearing classification among the multiple probabilities being above a particular threshold.
7. The apparatus according to claim 5 , wherein the word or expression frequently used prior to, within, or following an intent bearing utterance being provided to the classifier manually through rules or automatically identified in a data driven manner.
8. A non-transitory computer-readable medium for identifying at least one intent-bearing utterance in a conversation between a human agent associated with a call center, and a caller, the non-transitory computer-readable medium includes computer code stored thereon, the computer code when executed by a processor causes an apparatus to perform at least the following:
determining a plurality of features for each utterance among a set of utterances of the conversation between the human agent and the caller, the features including one of an identity associated with the utterance between the human agent and the caller, a position of the utterance within the conversation, and the word located before or after the utterance within the conversation;
classifying each utterance among the set of utterances, using a classifier, as an intent-bearing classification or a non-intent-bearing classification by generating multiple state sequences for each set of utterances, each state sequence including a plurality of states, each of the plurality of states corresponding to a respective utterance among the set of utterances and each of the plurality of states representing a probability that the respective utterance is intent-bearing, each utterance being associated with respective probabilities of multiple state sequences;
generating a ranking of each utterance based on a combination of the respective probabilities of the multiple state sequences, the ranking being a relative measure that the respective utterance is intent-bearing; and
selecting an utterance as being an intent-bearing utterance representing an intent of the conversation based on the ranking.
9. The non-transitory computer-readable medium according to claim 8 , wherein in classifying each utterance of the set of utterances, the computer code when executed by a processor causes an apparatus to further:
wherein in selecting at least one utterance, the computer code when executed by a processor causes the apparatus to select the at least one utterance based on highest occurrence of intent classification among the probabilities being above a particular threshold.