Using feature aggregation to determine intent classifications for text transcripts

1 . A method comprising:

accessing, by an intent classification engine, a text transcript;

determining, by the intent classification engine, one or more features associated with the text transcript, the determining the one or more features comprising:

generating, based at least in part on the text transcript, a graph comprising a plurality of nodes interconnected by a plurality of edges; and

using the graph to determine at least one of the one or more features associated with the text transcript;

generating, by the intent classification engine based on the one or more features, an aggregate embedding vector; and

providing, by the intent classification engine, the aggregate embedding vector as an input to a trained model configured to output an intent classification,

wherein:

the determining the one or more features comprises determining, by the intent classification engine based on the graph, a co-occurrence ranking associated with one or more local intent classifications represented in the graph; and

the one or more features comprises the co-occurrence ranking.

2 . The method of claim 1 , wherein the one or more features comprises the one or more local intent classifications and one or more of a sentence type, an emotion, an intent recency, an intent frequency, a sequence weight, a geo-location intent, or a previous intent.

3 . The method of claim 1 , wherein the determining the one or more features comprises:

providing, by the intent classification engine, one or more portions of the text transcript as inputs to a second trained model configured to output the one or more local intent classifications;

wherein the one or more features comprises the one or more local intent classifications.

4 . The method of claim 3 , wherein each of the one or more portions comprises one or more of a sentence or a phrase within the text transcript.

5 . The method of claim 3 , wherein the second trained model comprises a multi-layer neural network.

6 . The method of claim 1 , wherein each node of the graph corresponds to a local intent classification associated with the text transcript and each edge of the graph corresponds to a co-occurrence frequency between two respective nodes.

7 . The method of claim 6 , where each co-occurrence frequency is based on previous text transcripts associated with the two respective nodes.

8 . The method of claim 1 , wherein the trained model comprises a multi-layer neural network.

9 . The method of claim 1 , wherein the text transcript is associated with a customer support interaction with an agent.

10 . The method of claim 1 , wherein the accessing the text transcript occurs in real-time while the text transcript is being generated.

11 . The method of claim 1 , wherein the accessing the text transcript occurs in real-time after a threshold number of sentences within the text transcript have been generated.

12 . A system comprising:

a memory storing instructions; and

one or more processors communicatively coupled to the memory and configured to execute the instructions to perform a process comprising:

accessing a text transcript;

determining one or more features associated with the text transcript, the determining the one or more features comprising:

generating, based at least in part on the text transcript, a graph comprising a plurality of nodes interconnected by a plurality of edges; and

using the graph to determine at least one of the one or more features associated with the text transcript;

generating, based on the one or more features, an aggregate embedding vector; and

providing the aggregate embedding vector as an input to a trained model configured to output an intent classification,

wherein:

the determining the one or more features comprises determining, based on the graph, a co-occurrence ranking associated with one or more local intent classifications represented in the graph; and

the one or more features comprises the co-occurrence ranking.

13 . The system of claim 12 , wherein the determining the one or more features comprises:

providing one or more portions of the text transcript as inputs to a second trained model configured to output the one or more local intent classifications;

wherein the one or more features comprises the one or more local intent classifications.

14 . The system of claim 13 , wherein each of the one or more portions comprises one or more of a sentence or a phrase within the text transcript.

15 . The system of claim 12 , wherein each node of the graph corresponds to a local intent classification associated with the text transcript and each edge of the graph corresponds to a co-occurrence frequency between two respective nodes.

16 . The system of claim 12 , wherein the trained model comprises a multi-layer neural network.

17 . The system of claim 12 , wherein the accessing the text transcript occurs in real-time after a threshold number of sentences within the text transcript have been generated.

18 . A non-transitory computer-readable medium storing instructions that, when executed, direct a processor of a computing device to perform a process comprising:

accessing a text transcript;

determining one or more features associated with the text transcript, the determining the one or more features comprising:

generating, based at least in part on the text transcript, a graph comprising a plurality of nodes interconnected by a plurality of edges; and

using the graph to determine at least one of the one or more features associated with the text transcript;

generating, based on the one or more features, an aggregate embedding vector; and

providing the aggregate embedding vector as an input to a trained model configured to output an intent classification,

wherein:

the determining the one or more features comprises determining, based on the graph, a co-occurrence ranking associated with one or more local intent classifications represented in the graph; and

the one or more features comprises the co-occurrence ranking.

19 . The non-transitory computer-readable medium of claim 18 , wherein the determining the one or more features comprises:

providing one or more portions of the text transcript as inputs to a second trained model configured to output the one or more local intent classifications;

wherein the one or more features comprises the one or more local intent classifications.

20 . The non-transitory computer-readable medium of claim 19 , wherein each of the one or more portions comprises one or more of a sentence or a phrase within the text transcript.