Data-driven taxonomy for annotation resolution
An intent confusion evaluation engine receives conversation data corresponding to conversations between customers and agents. The engine evaluates annotations in the conversation data corresponding to intents identified from messages exchanged between customers and agents to determine levels of confusion amongst different intents. Based on these levels of confusion, the engine creates a graphical representation that illustrates the various intents and the level of confusion between different pairings of intents for the set of conversations. If an update is provided to the annotations, the graphical representation is updated dynamically and in real-time to provide updated levels of confusion amongst the various intents in accordance with the update.
1. A computer-implemented method comprising:
receiving ongoing conversation data that includes exchanged messages;
applying a machine-learning model to the ongoing conversation data to generate a set of annotations for the messages, wherein an annotation of the set of annotations corresponds to a particular message, and wherein the annotation of the particular message includes a set of confidence scores corresponding to classifications;
generating an estimated amount of confusion present in the classifications by analyzing the annotation and the set of confidence scores, wherein the estimated amount of confusion represents a level of confusion between a pair of the classifications;
dynamically generating in real-time a graphical representation that includes a set of nodes and a set of edges, wherein two nodes represent the pair of classifications, wherein an edge connects the two nodes and represents the estimated amount of confusion between the pair of classifications, and wherein the graphical representation is generated as the ongoing conversation data continues to be received;
receiving input corresponding to an interaction with the graphical representation;
modifying a classification of the pair of the classifications according to the interaction, wherein modifying includes decreasing the estimated amount of confusion between the pair of classifications;
updating the machine-learning model based on the modified classification; and
applying the updated machine-learning model to the ongoing conversation data to generate an updated set of annotations for the messages, wherein an updated annotation includes an updated confidence score corresponding to the modified classification.
2. The computer-implemented method of claim 1 , wherein the ongoing conversation data include text data, audio data, or video data.
3. The computer-implemented method of claim 1 , wherein the interaction includes merging two classifications of the classifications into a single merged classification.
4. The computer-implemented method of claim 1 , wherein the machine-learning model is trained using sample conversation data and corresponding annotations.
5. The computer-implemented method of claim 1 , further comprising:
dynamically modifying the graphical representation based on the modified classification, wherein the graphical representation is dynamically modified to include an updated set of nodes and an updated set of edges, wherein two updated nodes include the modified classification, and wherein an updated edge represents a re-estimated amount of confusion between the modified classification and another classification.
6. The computer-implemented method of claim 1 , wherein generating the estimated amount of confusion includes determining a conditional probability associated with the pair of classifications, and wherein the conditional probability is determined based on a probability of a first classification of the pair of the classifications being selected after a previous selection of a second classification of the pair of classification.
7. A system comprising:
one or more processors; and
memory storing thereon instructions that, as a result of being executed by the one or more processors, cause the system to perform operations comprising:
receiving ongoing conversation data that includes exchanged messages;
applying a machine-learning model to the ongoing conversation data to generate a set of annotations for the messages, wherein an annotation of the set of annotations corresponds to a particular message, and wherein the annotation of the particular message includes a set of confidence scores corresponding to classifications;
generating an estimated amount of confusion present in the classifications by analyzing the annotation and the set of confidence scores, wherein the estimated amount of confusion represents a level of confusion between a pair of the classifications;
dynamically generating in real-time a graphical representation that includes a set of nodes and a set of edges, wherein two nodes represent the pair of classifications, wherein an edge connects the two nodes and represents the estimated amount of confusion between the pair of classifications, and wherein the graphical representation is generated as the ongoing conversation data continues to be received;
receiving input corresponding to an interaction with the graphical representation;
modifying a classification of the pair of the classifications according to the interaction, wherein modifying includes decreasing the estimated amount of confusion between the pair of classifications;
updating the machine-learning model based on the modified classification; and
applying the updated machine-learning model to the ongoing conversation data to generate an updated set of annotations for the messages, wherein an updated annotation includes an updated confidence score corresponding to the modified classification.
8. The system of claim 7 , wherein the ongoing conversation data include text data, audio data, or video data.
9. The system of claim 7 , wherein the interaction includes merging two classifications of the classifications into a single merged classification.
10. The system of claim 7 , wherein the machine-learning model is trained using sample conversation data and corresponding annotations.
11. The system of claim 7 , wherein the instructions further cause the system to perform operations comprising:
dynamically modifying the graphical representation based on the modified classification, wherein the graphical representation is dynamically modified to include an updated set of nodes and an updated set of edges, wherein two updated nodes include the modified classification, and wherein an updated edge represents a re-estimated amount of confusion between the modified classification and another classification.
12. The system of claim 7 , wherein generating the estimated amount of confusion includes determining a conditional probability associated with the pair of classifications, and wherein the conditional probability is determined based on a probability of a first classification of the pair of the classifications being selected after a previous selection of a second classification of the pair of classification.
13. A non-transitory, computer-readable storage medium storing thereon executable instructions that, as a result of being executed by one or more processors of a computer system, cause the computer system to perform operations comprising:
receiving ongoing conversation data that includes exchanged messages;
applying a machine-learning model to the ongoing conversation data to generate a set of annotations for the messages, wherein an annotation of the set of annotations corresponds to a particular message, and wherein the annotation of the particular message includes a set of confidence scores corresponding to classifications;
generating an estimated amount of confusion present in the classifications by analyzing the annotation and the set of confidence scores, wherein the estimated amount of confusion represents a level of confusion between a pair of the classifications;
dynamically generating in real-time a graphical representation that includes a set of nodes and a set of edges, wherein two nodes represent the pair of classifications, wherein an edge connects the two nodes and represents the estimated amount of confusion between the pair of classifications, and wherein the graphical representation is generated as the ongoing conversation data continues to be received;
receiving input corresponding to an interaction with the graphical representation;
modifying a classification of the pair of the classifications according to the interaction, wherein modifying includes decreasing the estimated amount of confusion between the pair of classifications;
updating the machine-learning model based on the modified classification; and
applying the updated machine-learning model to the ongoing conversation data to generate an updated set of annotations for the messages, wherein an updated annotation includes an updated confidence score corresponding to the modified classification.
14. The non-transitory, computer-readable storage medium of claim 13 , wherein the ongoing conversation data include text data, audio data, or video data.
15. The non-transitory, computer-readable storage medium of claim 13 , wherein the interaction includes merging two classifications of the classifications into a single merged classification.
16. The non-transitory, computer-readable storage medium of claim 13 , wherein the machine-learning model is trained using sample conversation data and corresponding annotations.
17. The non-transitory, computer-readable storage medium of claim 13 , wherein the executable instructions further cause the computer system to perform operations comprising:
dynamically modifying the graphical representation based on the modified classification, wherein the graphical representation is dynamically modified to include an updated set of nodes and an updated set of edges, wherein two updated nodes include the modified classification, and wherein an updated edge represents a re-estimated amount of confusion between the modified classification and another classification.
18. The non-transitory, computer-readable storage medium of claim 13 , wherein generating the estimated amount of confusion includes determining a conditional probability associated with the pair of classifications, and wherein the conditional probability is determined based on a probability of a first classification of the pair of the classifications being selected after a previous selection of a second classification of the pair of classification.