Language-model-based data augmentation method for textual classification tasks with little data
Embodiments of the present systems and methods may provide techniques for augmenting textual data that may be used for textual classification tasks. Embodiments of such techniques may provide the capability to synthesize labeled data to improve text classification tasks. Embodiments may be specifically useful when only a small amount of data is available, and provide improved performance in such cases. For example, in an embodiment, a method implemented in a computer system may comprise a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor, and the method may comprise fine-tuning a language model using a training dataset, synthesizing a plurality of samples using the fine-tuned language model, filtering the plurality of synthesized samples, and generating an augmented training dataset comprising the training dataset and the filtered plurality of synthesized sentences.
1. A method implemented in a computer system comprising a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor, the method comprising:
fine-tuning a language model using a training dataset, wherein fine-tuning of the language model comprises:
concatenating sentences in the training dataset according to:
U*=y 1 SEP x i EOS y 2 SEP x 2 EOS y 3 . . . y n SEP x n EOS
wherein SEP is an auxiliary token that separates between a class label and a corresponding sentence and EOS is a token that terminates a sentence and separates it from a label that follows;
synthesizing a plurality of samples using the fine-tuned language model, wherein synthesizing a plurality of samples comprises: generating a set of new labeled sentences starting from “y SEP”;
filtering the plurality of synthesized samples; and
generating an augmented training dataset comprising the training dataset and the filtered plurality of synthesized sentences.
2. The method of claim 1 , wherein the training dataset comprises a small quantity of training data.
3. The method of claim 2 , wherein the language model is pre-trained on a large quantity of training data.
4. The method of claim 3 , wherein the fine-tuning comprises adapting the language model to the linguistic domain of the training dataset.
5. The method of claim 4 , wherein the plurality of synthesized samples is filtered using a classifier model trained using the training dataset to score the quality of the synthesized samples.
6. The method of claim 4 , wherein the plurality of synthesized samples is filtered using a filtering method selected from a sentence length, vocabulary constraints, plausibility, and human filtering.
7. A system comprising a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor to perform:
fine-tuning a language model using a training dataset, wherein fine-tuning of the language model comprises:
concatenating sentences in the training dataset according to:
U*=y 1 SEP x i EOS y 2 SEP x 2 EOS y 3 . . . y n SEP x n EOS
wherein SEP is an auxiliary token that separates between a class label and a corresponding sentence and EOS is a token that terminates a sentence and separates it from a label that follows;
synthesizing a plurality of samples using the fine-tuned language model, wherein synthesizing a plurality of samples comprises: generating a set of new labeled sentences starting from “y SEP”;
filtering the plurality of synthesized samples; and
generating an augmented training dataset comprising the training dataset and the filtered plurality of synthesized sentences.
8. The system of claim 7 , wherein the training dataset comprises a small quantity of training data.
9. The system of claim 8 , wherein the language model is pre-trained on a large quantity of training data.
10. The system of claim 9 , wherein the fine-tuning comprises adapting the language model to the linguistic domain of the training dataset.
11. The system of claim 10 , wherein the plurality of synthesized samples is filtered using a classifier model trained using the training dataset to score the quality of the synthesized samples.
12. The system of claim 10 , wherein the plurality of synthesized samples is filtered using a filtering method selected from a sentence length, vocabulary constraints, plausibility, and human filtering.
13. A computer program product comprising a non-transitory computer readable storage having program instructions embodied therewith, the program instructions executable by a computer, to cause the computer to perform a method comprising:
fine-tuning a language model using a training dataset, wherein fine-tuning of the language model comprises:
concatenating sentences in the training dataset according to:
U*=y 1 SEP x i EOS y 2 SEP x 2 EOS y 3 . . . y n SEP x n EOS
wherein SEP is an auxiliary token that separates between a class label and a corresponding sentence and EOS is a token that terminates a sentence and separates it from a label that follows;
synthesizing a plurality of samples using the fine-tuned language model, wherein synthesizing a plurality of samples comprises: generating a set of new labeled sentences starting from “y SEP”;
filtering the plurality of synthesized samples; and
generating an augmented training dataset comprising the training dataset and the filtered plurality of synthesized sentences.
14. The computer program product of claim 13 , wherein the training dataset comprises a small quantity of training data.
15. The computer program product of claim 14 , wherein the language model is pre-trained on a large quantity of training data.
16. The computer program product of claim 15 , wherein the fine-tuning comprises adapting the language model to the linguistic domain of the training dataset.
17. The computer program product of claim 16 , wherein the plurality of synthesized samples is filtered using a filtering method selected from a classifier model trained using the training dataset to score the quality of the synthesized samples, a sentence length, vocabulary constraints, plausibility, and human filtering.