Generating summary content tuned to a target characteristic using a word generation model
Systems and methods for generating a tuned summary using a word generation model. An example method includes receiving, at a decoder of the word generation model, a training data learned subspace representation of training data. The method also includes identifying tunable linguistic characteristics of the word generation model and training the decoder to output a training tuned summary of the training data learned subspace representation based on at least one of the tunable linguistic characteristics. The method further includes receiving an input text and a target characteristic token, and generating, by the trained decoder of the word generation model, each word of a tuned summary of the input text from a learned subspace representation and from feedback about preceding words of the tuned summary, wherein the tuned summary is tuned to target characteristics represented by the target characteristic token.
1. A method for generating a tuned summary using a word generation model, wherein the method includes one or more processing devices performing operations comprising:
receiving, at a decoder of the word generation model, a training data learned subspace representation of training data;
identifying tunable linguistic characteristics of the word generation model;
training the decoder to output a training tuned summary of the training data learned subspace representation based on at least one of the tunable linguistic characteristics;
receiving an input text and a target characteristic token; and
generating, by the trained decoder of the word generation model, each word of a tuned summary of the input text from a learned subspace representation and from a feedback about preceding words of the tuned summary, wherein the tuned summary is tuned to target characteristics represented by the target characteristic token.
2. The method of claim 1 , wherein the one or more processing devices perform operations further comprising training the word generation model, wherein training the word generation model enables tuning of the tuned summary to a set of structural target characteristics, a set of content based target characteristics, a set of qualitative target characteristics, a set of linguistic target characteristics, or any combination thereof.
3. The method of claim 1 , wherein the one or more processing devices perform operations further comprising:
prepending a training token to a set of ground truth text to generate training data;
receiving the training data at the word generation model; and
training the word generation model using the training data, the training token, and a ground truth summary.
4. The method of claim 1 , wherein the one or more processing devices perform operations further comprising:
receiving, at an encoder, training data;
modifying a focus of attention at an attention distribution block of the word generation model using a hyperparameter; and
training the word generation model using end-to-end training with the training data and the hyperparameter.
5. The method of claim 1 , wherein the one or more processing devices perform operations further comprising:
receiving, at an encoder, training data;
identifying word generation model loss comprising log likelihood loss and return loss; and
training the word generation model using end-to-end training to reduce the word generation model loss.
6. The method of claim 1 , wherein training the decoder comprises using an optimum multiplicative factor to output the training tuned summary of the training data learned subspace representation.
7. The method of claim 1 , wherein the one or more processing devices perform operations further comprising:
increasing focus, by an attention distribution block, on portions of the input text that are relevant to the target characteristics represented by the target characteristic token prior to generating the tuned summary.
8. The method of claim 1 , wherein the one or more processing devices perform operations further comprising:
generating the learned subspace representation of the input text and the target characteristic token associated with the input text by applying an encoder of the word generation model to the input text and the target characteristic token.
9. A computing system comprising:
means for receiving, at a decoder of a word generation model, a training data learned subspace representation of training data;
means for identifying tunable linguistic characteristics of the word generation model;
means for training the decoder to output a training tuned summary of the training data learned subspace representation based on at least one of the tunable linguistic characteristics;
means for receiving an input text and a target characteristic token; and
means for generating, by the trained decoder of the word generation model, each word of a tuned summary of the input text from a learned subspace representation and from a feedback about preceding words of the tuned summary, wherein the tuned summary is tuned to target characteristics represented by the target characteristic token.
10. The computing system of claim 9 , further comprising means for training a word generation model, wherein training the word generation model enables tuning of the tuned summary to a set of structural target characteristics, a set of content based target characteristics, a set of qualitative target characteristics, a set of linguistic target characteristics, or any combination thereof.
11. The computing system of claim 9 , further comprising:
means for prepending a training token to a set of ground truth text to generate training data;
means for receiving the training data at a word generation model; and
means for training the word generation model using the training data, the training token, and a ground truth summary.
12. The computing system of claim 9 , further comprising:
means for receiving, at an encoder, training data;
means for modifying a focus of attention at an attention distribution block of a word generation model using a hyperparameter; and
means for training the word generation model using end-to-end training with the training data and the hyperparameter.
13. The computing system of claim 9 , further comprising:
means for receiving, at an encoder, training data;
means for identifying word generation model loss comprising log likelihood loss and return loss; and
means for training a word generation model using end-to-end training to reduce the word generation model loss.
14. The computing system of claim 9 , wherein the means for training the decoder comprises a means for using an optimum multiplicative factor to output the training tuned summary of the training data learned subspace representation.
15. The computing system of claim 9 , further comprising:
means for increasing focus, by an attention distribution block, on portions of the input text that are relevant to the target characteristics represented by the target characteristic token prior to generating the tuned summary.
16. A non-transitory computer-readable medium having instructions stored thereon, the instructions executable by a processing device to perform operations comprising:
receiving, at a decoder of a word generation model, a training data learned subspace representation of training data;
identifying tunable linguistic characteristics of the word generation model;
training the decoder to output a training tuned summary of the training data learned subspace representation based on at least one of the tunable linguistic characteristics;
receiving an input text and a target characteristic token; and
generating, by the trained decoder of the word generation model, each word of a tuned summary of the input text from a learned subspace representation and from a feedback about preceding words of the tuned summary, wherein the tuned summary is tuned to target characteristics represented by the target characteristic token.
17. The non-transitory computer-readable medium of claim 16 , the instructions further executable by the processing device perform operations comprising training a word generation model, wherein training the word generation model enables tuning of the tuned summary to a set of structural target characteristics, a set of content based target characteristics, a set of qualitative target characteristics, a set of linguistic target characteristics, or any combination thereof.
18. The non-transitory computer-readable medium of claim 16 , the instructions further executable by the processing device perform operations comprising:
prepending a training token to a set of ground truth text to generate training data;
receiving the training data at a word generation model; and
training the word generation model using the training data, the training token, and a ground truth summary.
19. The non-transitory computer-readable medium of claim 16 , the instructions further executable by the processing device perform operations comprising:
receiving, at an encoder, training data;
modifying a focus of attention at an attention distribution block of a word generation model using a hyperparameter; and
training the word generation model using end-to-end training with the training data and the hyperparameter.
20. The non-transitory computer-readable medium of claim 16 , wherein generating the tuned summary comprises generating the tuned summary in a word-by-word manner.