Semantic frame builder

1. A method for building semantic frames, the method further comprising:

receiving, at an entity Interactive Voice Response (IVR) system, an utterance, via a natural language process (NLP) pipeline, said utterance being transmitted from a network node to the entity IVR system;

extracting featurized data included in the utterance, said extracting comprising:

tokenizing the utterance into a plurality of tokens;

indexing the plurality of tokens, said indexing comprising assigning an index number to each token included in the plurality of tokens, said index number based on a position of each token within the utterance; and

extracting metadata for each token included in the plurality of tokens;

classifying each token, within the plurality of tokens using a plurality of semantic role classifiers, said classifying comprising:

selecting one or more tokens from the plurality of tokens for semantic role classification;

assigning a semantic role classifier to each of the one or more selected tokens;

for each of the one or more selected tokens, identifying a correlation value between each semantic role classifier, included in the plurality of semantic role classifiers, and a selected token;

for each of the one or more selected tokens, finalizing an associated semantic role classifier, said finalizing based on determining the associated semantic role classifier having a correlation value that is a highest correlation value; and,

for each of the one or more selected tokens, determining whether there is a dependency relationship with any of other tokens, included in the plurality of tokens, said tokens that are determined to have a dependency relationship being identified as dependent tokens, said determining comprising storing index numbers of the dependent tokens;

building a semantic frame for each selected token, said semantic frame comprising:

the index number of the selected token;

the semantic role classifier assigned to the selected token;

the correlation value of the selected token and the associated semantic role classifier; and

the index numbers of the one or more dependent tokens within the plurality of tokens.

2. The method of claim 1 , further comprising using a machine learning algorithm to build the semantic frame.

3. The method of claim 1 , further comprising building semantic frames for a plurality of utterances, said plurality of utterances occurring in a conversation.

4. The method of claim 3 , further comprising generating a summary of the conversation using the semantic frames, said generating the summary of the conversation using the dependent tokens to reduce redundant and non-advantageous tokens from the conversation.

5. The method of claim 4 , the method further comprising assigning a highest significance score to a most recent utterance.

6. The method of claim 1 , wherein the semantic frame, for each token, provides a plurality of layers to the utterance.

7. The method of claim 1 , further comprising using a dependency parser to identify a dependency between a selected token and any of the other one or more tokens included in the plurality of tokens.

8. The method of claim 1 , wherein the method uses a Java®-based model.

9. The method of claim 1 , further comprising predicting, for a first utterance, based on a plurality of intents and a plurality of semantic frames generated for the first utterance, a response to the first utterance.

10. A method for building a semantic frame, the method further comprising:

receiving, at an entity Interactive Voice Response (IVR) system, a plurality of utterances via a natural language process (NLP) pipeline, said plurality of utterances being transmitted from a plurality of nodes to the entity IVR system;

extracting featurized data included in the plurality of utterances, said extracting comprising:

tokenizing each of the plurality of utterances into a plurality of tokens;

indexing each plurality of tokens, said indexing comprising assigning an index number to each token included in the plurality of tokens, said index number based on a position of each token within the utterance; and

extracting metadata for each token included in the plurality of tokens;

classifying each token, within the plurality of tokens, using a plurality of semantic role classifiers, said classifying comprising:

selecting one or more tokens from the plurality of tokens for semantic role classification;

for each of the one or more selected tokens, identifying a correlation value between each of the plurality of semantic role classifiers and the selected token;

assigning a semantic role classifier to each of the one or more tokens, said assigned semantic role classifier having a correlation value that is a highest correlation value from a plurality of correlation values;

for each of the one or more selected tokens, determining a subset of tokens that includes a dependency relation with any of the other one or more tokens included in the plurality of tokens;

identifying each token included in the subset of tokens as dependent tokens; and

storing one or more index numbers for each dependent token;

building a semantic frame for each selected token, said semantic frame comprising;

the index number of the selected token;

the semantic role classifier assigned to the selected token;

the correlation value of the selected token and an associated semantic role classifier; and

the one or more index numbers of the one or more dependent tokens within the plurality of tokens.

11. An apparatus for semantic frame building, the apparatus further comprising:

an entity Interactive Voice Response (IVR) system, said entity IVR linked to a natural language process (NLP) pipeline, said entity IVR operable to receive an utterance from a network node via the NLP pipeline, the IVR comprising;

a data featurizer, said data featurizer configured to:

tokenize the utterance into a plurality of tokens;

index the plurality of tokens by assigning an index number to each token included in the plurality of tokens based on a position of each token within the utterance; and

extract metadata for each token included in the plurality of tokens;

a semantic role classifier, said semantic role classifier operable to:

select one for more tokens from the plurality of tokens for semantic role classification;

for each of the one or more selected tokens, identify a correlation value between a semantic role classifier selected from a plurality of semantic role classifiers and a selected token;

assign a semantic role classifier to each of the one or more selected tokens based on determination of the semantic role classifier correlating, above a predetermined correlation level, to the one or more selected tokens; and

for each of the one or more selected tokens, determine whether there is a dependency relation with any of the other one or more tokens included in the plurality of tokens, said tokens that are determined to have a dependency relation being identified as dependent tokens; and

store index numbers of the one or more dependent tokens;

a semantic frame builder said semantic frame builder configured to build a semantic frame for each token, said semantic frame comprising:

the index number of the selected token;

the semantic role classifier assigned to the selected token;

the correlation value of the selected token and an associated semantic role classifier; and

the index numbers of the one or more dependent tokens within the plurality of tokens.

12. The apparatus of claim 11 , the apparatus further comprising a machine learning algorithm, said machine learning algorithm operable to execute the data featurizer, semantic role classifier and semantic frame builder, said machine learning algorithm residing on a hardware processor and operating in tandem with hardware memory.

13. The apparatus of claim 12 , wherein the machine learning algorithm is further configured to predict for a first utterance, based on a plurality of intents and a plurality of semantic frames generated for the first utterance, a response to the first utterance.

14. The apparatus of claim 11 , further configured to build semantic frames for a plurality of utterances, said plurality of utterances occurring in a conversation.

15. The apparatus of claim 14 , the apparatus further configured to generate a summary of the conversation using the semantic frames using the identified dependent tokens to reduce redundant and non-advantageous tokens from the conversation.

16. The apparatus of claim 15 , the apparatus further configured to assign a highest significance score to a most recent utterance.

17. The apparatus of claim 11 , wherein a semantic frame, for each token, provides a plurality of layers to the utterance.

18. The apparatus of claim 11 , further configured to use a dependency parser to identify a dependency between a selected token and another token included in the plurality of tokens.

19. The apparatus of claim 18 , the dependency parser further configured to identify dependencies and relationships between tokens, said dependency parser configured to:

analyze a plurality of tokens included in an utterance;

identify parts of speech of one or more tokens included in the plurality of tokens; and

based on the identified parts of speech, extract a set of relationships and dependencies between the one or more tokens included in the plurality of tokens.

20. The apparatus of claim 11 , wherein the semantic frame builder uses a Java®-based model.