Machine learning used to detect alignment and misalignment in conversation
Digitized media is received that records a conversation between individuals. Cues are extracted from the digitized media that indicate properties of the conversation. The cues are entered as training data into a machine learning module to create a trained machine learning model. The trained machine learning model is used in a processor to detect other misalignments in subsequent digitized conversations.
1. A method comprising:
receiving digitized media that represents a conversation between individuals;
extracting cues from the digitized media that indicate properties of the conversation;
entering the cues as training data into a machine learning module to create a trained machine learning model that detects misalignments in subsequent conversations between individuals, wherein misalignment is a lack of agreement and mutual understanding between the individuals of content and context of the content of the subsequent conversations; and
using the trained machine learning model in a processor to detect other misalignments in subsequent digitized conversations;
wherein the trained machine learning model detects misalignments in subsequent digitized conversations between individuals by determining a probability of each cue occurring in the subsequent digitized conversations under a normalcy model.
2. The method of claim 1 , wherein extracting the cues comprises:
extracting lower-level cues from the digitized media; and
constructing higher-level cues based on the lower level cues, wherein the higher-level cues comprise latent topics associated with the conversation.
3. The method of claim 2 , wherein constructing the higher-level cues comprises processing the lower-level cues using at least one of a latent Dirichlet allocation and a hidden Markov model.
4. The method of claim 1 , wherein the machine learning module comprises a support vector machine.
5. The method of claim 1 , wherein the cues comprise prosodic cues.
6. The method of claim 1 , wherein extracting the cues is performed by a deep neural network.
7. The method of claim 1 , wherein the conversation is a customer support session, and wherein using the trained machine learning model to detect the other misalignments comprises using the trained machine learning model to evaluate call performance of customer service agents based on the other misalignments.
8. The method of claim 7 , wherein evaluating the call performance comprises displaying a real time dashboard that displays call performance of an agent relative to a baseline.
9. The method of claim 7 , wherein evaluating the call performance comprises providing statistics of a plurality of agents in a call center, the statistics used to evaluate a call center policy.
10. The method of claim 1 , wherein the digitized media includes one or more of audio and video streams.
11. The method of claim 1 , wherein the training data further comprises any combination of the digitized media and an external source of training data.
12. The method of claim 1 , further comprising fine-tuning the machine learning module using any combination of the subsequent digitized conversations and human-supplied ground truth labels.
13. A system comprising:
a circuit operable to receive digitized media that represents a conversation between at least two people; and
a first processor coupled to the circuit and operable to:
extract cues from the digitized media that indicate properties of the conversation; and
enter the cues as training data into a machine learning module to create a trained machine learning model; and
a second processor configured to:
receive the trained machine learning model from the first processor; and
use the trained machine learning model in a processor to detect other misalignments in subsequent digitized conversations between individuals, wherein misalignment is a lack of agreement and mutual understanding between the individuals of content and context of the content of the subsequent digitized conversations;
wherein the trained machine learning model detects misalignments in subsequent digitized conversations between individuals by determining a probability of each cue occurring in the subsequent digitized conversations under a normalcy model.
14. The system of claim 13 , wherein extracting the cues comprises:
extracting lower-level cues from the digitized media; and
constructing higher-level cues based on the lower level cues, wherein the higher-level cues comprise latent topics associated with the conversation.
15. The system of claim 14 , wherein the digitized media comprises a textual transcript of the conversation, the lower level cues further comprising textual cues obtained from the textual transcript.
16. The system of claim 14 , wherein constructing the higher-level cues comprises processing the lower-level cues using at least one of a latent Dirichlet allocation and a hidden Markov model.
17. The system of claim 13 , wherein the training data further comprises any combination of the digitized media and an external source of training data.
18. The system of claim 13 , wherein the first processor is further operable to fine tune the machine learning module using any combination of the subsequent digitized conversations and human-supplied ground truth labels.
19. The system of claim 13 , wherein the conversation is a customer support session, and wherein using the trained machine learning model to detect the other misalignments comprises using the trained machine learning model to evaluate call performance of customer service agents based on the other misalignments.