Training data augmentation using gazetteers and perturbations to facilitate training named entity recognition models
Techniques are provided for augmenting training data using gazetteers and perturbations to facilitate training named entity recognition models. The training data can be augmented by generating additional utterances from original utterances in the training data and combining the generated additional utterances with the original utterances to form the augmented training data. The additional utterances can be generated by replacing the named entities in the original utterances with different named entities and/or perturbed versions of the named entities in the original utterances selected from a gazetteer. Gazetteers of named entities can be generated from the training data and expanded by searching a knowledge base and/or perturbing the named entities therein. The named entity recognition model can be trained using the augmented training data.
1. A computer-implemented method comprising:
accessing training data comprising a plurality of original utterances, wherein each original utterance of the plurality of original utterances comprises at least one named entity corresponding to a named entity category of a plurality of named entity categories;
accessing one or more gazetteers, wherein each gazetteer of the one or more gazetteers comprises a plurality of named entities extracted from the plurality of original utterances and a plurality of perturbed named entities derived from one or more named entities of the plurality of named entities;
generating a plurality of template utterances, wherein each template utterance of the plurality of template utterances comprises information from an original utterance of the plurality of original utterances and at least one placeholder identifier representing a named entity in the original utterance of the plurality of original utterances;
generating a plurality of additional utterances, wherein each additional utterance of the plurality of additional utterances comprises a template utterance of the plurality of template utterances populated with at least one named entity selected from a gazetteer of the one or more gazetteers;
augmenting the training data by adding the plurality of additional utterances to the plurality of original utterances; and
training a named entity recognition (NER) model with the augmented training data.
2. The computer-implemented method of claim 1 , wherein each gazetteer of the one or more gazetteers corresponds to a different named entity category of the plurality of named entity categories.
3. The computer-implemented method of claim 1 , wherein at least one gazetteer of the one or more gazetteers comprises a plurality of named entities retrieved from a source other than the training data.
4. The computer-implemented method of claim 3 , wherein the plurality of named entities retrieved from the source other than the training data comprises named entities retrieved using at least one of a pre-trained model and a query-based search.
5. The computer-implemented method of claim 4 , wherein the source other than the training data comprises at least one knowledge base.
6. The computer-implemented method of claim 1 , wherein the plurality of perturbed named entities derived from one or more named entities of the plurality of named entities comprises perturbed versions of named entities of the plurality of named entities.
7. The computer-implemented method of claim 6 , wherein a particular perturbed version of the perturbed versions of named entities of the plurality of named entities comprises a named entity having at least one typographical error.
8. The computer-implemented method of claim 1 , further comprising:
providing the trained NER model to a system, wherein the providing the trained NER model includes detecting and classifying named entities in utterances received by the system from a user.
9. A system comprising:
one or more processors; and
one or more non-transitory computer-readable media storing instructions which, when executed by the one or more processors, cause the system to perform operations comprising:
accessing training data comprising a plurality of original utterances, wherein each original utterance of the plurality of original utterances comprises at least one named entity corresponding to a named entity category of a plurality of named entity categories;
accessing one or more gazetteers, wherein each gazetteer of the one or more gazetteers comprises a plurality of named entities extracted from the plurality of original utterances and a plurality of perturbed named entities derived from one or more named entities of the plurality of named entities;
generating a plurality of template utterances, wherein each template utterance of the plurality of template utterances comprises information from an original utterance of the plurality of original utterances and at least one placeholder identifier representing a named entity in the original utterance of the plurality of original utterances;
generating a plurality of additional utterances, wherein each additional utterance of the plurality of additional utterances comprises a template utterance of the plurality of template utterances populated with at least one named entity selected from a gazetteer of the one or more gazetteers;
augmenting the training data by adding the plurality of additional utterances to the plurality of original utterances; and
training a named entity recognition (NER) model with the augmented training data.
10. The system of claim 9 , wherein each gazetteer of the one or more gazetteers corresponds to a different named entity category of the plurality of named entity categories.
11. The system of claim 9 , wherein at least one gazetteer of the one or more gazetteers comprises a plurality of named entities retrieved from a particular source other than the training data.
12. The system of claim 11 , wherein the plurality of named entities retrieved from the particular source other than the training data comprises named entities retrieved using at least one of a pre-trained model and a query-based search.
13. The system of claim 12 , wherein the particular source other than the training data comprises at least one knowledge base.
14. The system of claim 9 , wherein the plurality of perturbed named entities derived from one or more named entities of the plurality of named entities comprises perturbed versions of named entities of the plurality of named entities.
15. The system of claim 14 , wherein a particular perturbed version of the perturbed versions of named entities of the plurality of named entities comprises a named entity having at least one typographical error.
16. The system of claim 9 , the operations further comprising:
providing the trained NER model to a system, wherein providing the trained NER model includes detecting and classifying named entities in utterances received by the system in which the trained NER model is provided to from a user.
17. One or more non-transitory computer-readable media storing instructions which, when executed by one or more processors, cause a system to perform operations comprising:
accessing training data comprising a plurality of original utterances, wherein each original utterance of the plurality of original utterances comprises at least one named entity corresponding to a named entity category of a plurality of named entity categories;
accessing one or more gazetteers, wherein each gazetteer of the one or more gazetteers comprises a plurality of named entities extracted from the plurality of original utterances and a plurality of perturbed named entities derived from one or more named entities of the plurality of named entities;
generating a plurality of template utterances, wherein each template utterance of the plurality of template utterances comprises information from an original utterance of the plurality of original utterances and at least one placeholder identifier representing a named entity in the original utterance of the plurality of original utterances;
generating a plurality of additional utterances, wherein each additional utterance of the plurality of additional utterances comprises a template utterance of the plurality of template utterances populated with at least one named entity selected from a gazetteer of the one or more gazetteers;
augmenting the training data by adding the plurality of additional utterances to the plurality of original utterances; and
training a named entity recognition (NER) model with the augmented training data.
18. The one or more non-transitory computer-readable media of claim 17 , wherein each gazetteer of the one or more gazetteers corresponds to a different named entity category of the plurality of named entity categories.
19. The one or more non-transitory computer-readable media of claim 17 , wherein at least one gazetteer of the one or more gazetteers comprises a plurality of named entities retrieved from a particular source other than the training data.
20. The one or more non-transitory computer-readable media of claim 17 , wherein the plurality of perturbed named entities derived from one or more named entities of the plurality of named entities comprises perturbed versions of named entities of the plurality of named entities.