Automated system and method to prioritize language model and ontology expansion and pruning

1. A computer-implemented method of automatically removing terms from a language model based on use, the method comprising:

ingesting language data from internet-based content streams, wherein the internet-based content streams are specific to a domain of an intelligent virtual assistant (IVA);

processing the language data using a language model toolkit to extract vocabulary from the language data, the extracted vocabulary comprising terms;

identifying select terms by counting the terms in the extracted vocabulary over a sliding window;

comparing the select terms from the extracted vocabulary to terms stored in a language model specific to the IVA to determine if the select terms have a match to the terms in the language model;

determining usage of the select terms in context with respect to the language data using an embedding model;

for any of the select terms that have a match in the language model, determining a frequency of use of the select term in the ingested language data in its associated usage in context;

automatically deleting one or more first terms of the select terms from a training example for the language model without further human intervention based on the frequency of use of the one or more first terms falls below a predetermined threshold;

recompiling the language model based on the training example after deleting the one or more first terms; and

adopting the recompiled language model into the IVA.

2. The method of claim 1 , further comprising adding one or more second terms the select terms to the training example if the frequency of use of the one or more second terms meets a second predetermined threshold.

3. The method of claim 1 , wherein the predetermined threshold is determined automatically based on predetermined values stored in a digital storage device.

4. A computer program product for automatically updating a language model for an intelligent virtual assistant (IVA), the computer program product comprising computer executable instructions embodied in a non-transitory computer readable medium that, when executed by one or more processors, cause the one or more processors to:

ingest language data from internet-based content streams, wherein the internet-based content streams are specific to a domain of the IVA;

process the language data using a language model to extract vocabulary from the language data, the extracted vocabulary comprising terms;

identify select terms by counting the terms in the extracted vocabulary over a sliding window;

compare the select terms from the extracted vocabulary to terms stored in a language model specific to the IVA to determine if the select terms have a match to the terms in the language model;

determine usage of the select terms in context with respect to the language data using an embedding model;

for any of the select terms that have a match in the language model, determine a frequency of use of the select term in the language data in its associated usage in context;

automatically delete one or more first terms of the select terms from a training example for the language model without further human intervention based on the frequency of use of the one or more first terms falls below a predetermined threshold;

recompile the language model based on the training example after deleting the one or more first terms; and

adopting the recompiled language model into the IVA.

5. The computer program product of claim 4 , further comprising additional instructions embodied in the non-transitory computer readable medium that, when executed by the one or more processors, cause the one or more processors to to add one or more second terms the select terms in the known context to the training example if the frequency of use of the one or more second terms meets a second predetermined threshold.

6. The computer program product of claim 4 , wherein the predetermined threshold is determined automatically based on predetermined values stored in a digital storage device.

7. A system for selecting actions to perform for an entity based on content items, the system comprising:

at least one processor; and

a memory storing instructions that, when executed by the at least one processor, cause the at least one processor to:

ingest language data from internet-based content streams, wherein the internet-based content streams are specific to a domain of an intelligent virtual assistant (IVA);

process the language data using a language model to extract vocabulary from the language data, the extracted vocabulary comprising terms;

identify select terms by counting the terms in the extracted vocabulary over a sliding window;

compare the select terms from the extracted vocabulary to terms stored in a language model specific to the IVA to determine if the select terms have a match to the terms in the language model;

determine usage of the select terms in context with respect to the language data using an embedding model;

for any of the select terms that have a match in the language model, determine a frequency of use of the select term in the language data in its associated context;

automatically delete first terms one or more of the select terms from a training example for the language model without further human intervention based on the frequency of use of the one or more first terms falls below a predetermined threshold

recompile the language model based on the training example after deleting the one or more first terms; and

adopt the recompiled language model into the IVA.

8. The system of claim 7 , the memory storing additional instructions that, when executed by the at least one processor, cause the at least one processor to adding one or more second terms the select terms in the known context to the training example if the frequency of use of the one or more second terms meets a second predetermined threshold.

9. The system of claim 7 , wherein the predetermined threshold is determined automatically based on predetermined values stored in a digital storage device.