Systems and methods for automatic extraction of classification training data
View Patent ↗A method for training a multi-class classification model includes receiving training data corresponding to a plurality of classes. For each class in the plurality of classes, the method includes training a binary classification model configured to determine whether or not an observation of training data belongs to the class and for each observation of training data identified as belonging to the class, extracting one or more class identification features from the observation of training data based on activations of an intermediate attention layer in the binary classification model. A multi-class classification model is trained using the class identification features extracted for each of the plurality of classes.
1. A computer-implemented method for training a multi-class classification model, comprising:
receiving, by a processor, training data corresponding to a plurality of classes, wherein the training data comprises a first corpus and a second corpus, wherein the first corpus is constructed such that each observation of training data is relevant to a respective class of the plurality of classes, and wherein the second corpus includes collected real world data;
for each class in the plurality of classes:
training, by the processor, a binary classification model configured to determine whether or not an observation of training data of the first corpus belongs to a class based on activations of an intermediate attention layer, wherein the binary classification model comprises a plurality of layers including an input layer and the intermediate attention layer, the input layer configured to divide an input text document into a plurality of sub-features and to evaluate a relative importance of each of a plurality of sub-features in the input text document, and the intermediate attention layer including a plurality of nodes corresponding to the plurality of sub-features of the input text document, wherein the binary classification model uses the relative importance to identify whether the input text document belongs to the class; and
extracting one or more class identification features from the observation of training data of the second corpus based on activations of the intermediate attention layer in the binary classification model; and
training, by the processor, the multi-class classification model using the class identification features extracted for each of the plurality of classes.
2. The method of claim 1 , wherein the multi-class classification model is a text classification model, and the training data comprises a plurality of text documents.
3. The method of claim 2 , wherein each class in the plurality of classes corresponds to an intended purpose of a text document.
4. The method of claim 2 , wherein the class identification features correspond to sentences extracted from the plurality of text documents.
5. The method of claim 2 , wherein extracting the class identification features comprises dividing each of the training data text documents into a plurality of sentences.
6. The method of claim 5 , wherein extracting the class identification features comprises evaluating a metric for each sentence in the plurality of text documents.
7. The method of claim 1 , wherein each class identification feature is extracted based on whether or not an activation weight of the intermediate attention layer is higher than a predefined threshold.
8. The method of claim 1 , wherein extracting the class identification features includes validating each feature using the corresponding binary classification model.
9. The method of claim 1 , wherein the binary classification model comprises the intermediate attention layer followed by a fully connected layer.
10. The method of claim 1 , further comprising labelling one or more observation of training data which do not belong to any of the plurality of classes.
11. A processing apparatus comprising a processor configured to execute a method comprising the steps of:
receiving, by the processor, training data corresponding to a plurality of classes, wherein the training data comprises a first corpus and a second corpus, wherein the first corpus is constructed such that each observation of training data is relevant to a respective class of the plurality of classes, and wherein the second corpus includes collected real world data; and, for each class in the plurality of classes:
training, by the processor, a binary classification model configured to determine whether or not an observation of training data of the first corpus belongs to a class based on activations of an intermediate attention layer, wherein the binary classification model comprises a plurality of layers including an input layer and the intermediate attention layer, the input layer configured to divide an input text document into a plurality of sub-features and to evaluate a relative importance of each of a plurality of sub-features in the input text document, and the intermediate attention layer including a plurality of nodes corresponding to the plurality of sub-features of the input text document, wherein the binary classification model uses the relative importance to identify whether the input text document belongs to the class; and
extracting one or more class identification features from the observation of training data of the second corpus based on activations of the intermediate attention layer in the binary classification model; and
training, by the processor, a multi-class classification model using the class identification features extracted for each of the plurality of classes.
12. The processing apparatus of claim 11 , wherein the multi-class classification model is a text classification model, and the training data comprises a plurality of text documents.
13. The processing apparatus of claim 12 , wherein each class in the plurality of classes corresponds to an intended purpose of a text document.
14. The processing apparatus of claim 12 , wherein the class identification features correspond to sentences extracted from the plurality of training text documents.
15. The processing apparatus of claim 12 , wherein extracting the class identification features comprises dividing each of the training data text documents into a plurality of sentences, and wherein extracting the class identification features comprises evaluating a metric for each sentence in the plurality of text documents.
16. A non-transitory, computer-readable medium configured to store instructions which, when executed by a processor, causes the processor to execute a method comprising the steps of:
receiving, by the processor, training data corresponding to a plurality of classes, wherein the training data comprises a first corpus and a second corpus, wherein the first corpus is constructed such that each observation of training data is relevant to a respective class of the plurality of classes, and wherein the second corpus includes collected real world data; and, for each class in the plurality of classes:
training, by the processor, a binary classification model configured to determine whether or not an observation of training data of the first corpus belongs to a class based on activations of an intermediate attention layer, wherein the binary classification model comprises a plurality of layers including an input layer and the intermediate attention layer, the input layer configured to divide an input text document into a plurality of sub-features and to evaluate a relative importance of each of a plurality of sub-features in the input text document, and the intermediate attention layer including a plurality of nodes corresponding to the plurality of sub-features of the input text document, wherein the binary classification model uses the relative importance to identify whether the input text document belongs to the class; and
extracting one or more class identification features from the observation of training data of the second corpus based on activations of the intermediate attention layer in the binary classification model; and
training, by the processor, a multi-class classification model using the class identification features extracted for each of the plurality of classes.
17. The non-transitory, computer-readable medium of claim 16 , wherein each class identification feature is extracted based on whether or not an activation weight of the intermediate attention layer is higher than a predefined threshold.
18. The non-transitory, computer-readable medium of claim 16 , wherein extracting the class identification features includes validating each feature using the corresponding binary classification model.
19. The non-transitory, computer-readable medium of claim 16 , wherein the binary classification model comprises the intermediate attention layer followed by a fully connected layer.
20. The non-transitory, computer-readable medium of claim 16 , wherein the method further comprising labelling one or more observation of training data which do not belong to any of the plurality of classes.