Back to results
US 12,608,403 B2Granted
Extraction machine learning framework
AssigneeSchlumberger Technology Corporation
InventorsOmkar Anil Gune, Purnaprajna Raghavendra Mangsuli, Michael Antony Cleminson
Life of this patent
Estimating the term…
Abstract
A method implements extraction using a machine learning framework. The method includes extracting a set of features from a table. The method further includes applying a feature extraction model to the set of features to generate a pseudo sentence. The method further includes applying a classifier model to the pseudo sentence to generate a label prediction for the table. The classifier model is trained with an augmented training pseudo sentence created by one or more of sampling and shuffling. The method further includes updating a label of the table with the label prediction.
Claims18
A method, comprising:
- extracting a set of features from a table;
- applying a feature extraction model to the set of features to generate a pseudo sentence;
- applying a classifier model to the pseudo sentence to generate a label prediction for the table, the classifier model being trained with an augmented training pseudo sentence, the classifier model comprising a neural network model comprising a plurality of layers, the plurality of layers including a plurality of vectorization layers comprising at least one of:
- a language model comprising at least an embedding layer and a transformer layer or an encoder comprising at least one or more hidden layers, to process the pseudo sentence and generate at least one classification vector, the classifier model being trained with the augmented training pseudo sentence created by:
- extracting a set of training features from a training table;
- sampling the set of training features to create a set of sampled features;
- creating a set of augmented features by:
- shuffling one or more of the set of training features or the set of sampled features;
- and
- dropping at least one of the one or more of the set of training features or the set of sampled features, the dropping being performed using weights identifying a likelihood that a feature is dropped;
- and
- applying the feature extraction model to the set of augmented features to create the augmented training pseudo sentence;
- and
- updating a label of the table with the label prediction.
11 more elements
A system, comprising:
- at least one processor;
- and
- an application, when executed by the at least one processor, performs:
- extracting a set of features from a table;
- applying a feature extraction model to the set of features to generate a pseudo sentence;
- applying a classifier model to the pseudo sentence to generate a label prediction for the table, the classifier model being trained with an augmented training pseudo sentence, the classifier model comprising a neural network model comprising a plurality of layers, the plurality of layers including a plurality of vectorization layers comprising at least one of:
- a language model comprising at least an embedding layer and a transformer layer or an encoder comprising at least one or more hidden layers, to process the pseudo sentence and generate at least one classification vector, the classifier model being trained with the augmented training pseudo sentence created by:
- extracting a set of training features from a training table;
- sampling the set of training features to create a set of sampled features;
- creating a set of augmented features by:
- shuffling one or more of the set of training features or the set of sampled features;
- and
- dropping at least one of the one or more of the set of training features or the set of sampled features, the dropping being performed using weights identifying a likelihood that a feature is dropped;
- and
- applying the feature extraction model to the set of augmented features to create the augmented training pseudo sentence;
- and
- updating a label of the table with the label prediction.
14 more elements
A non-transitory computer readable medium comprising instructions that, when executed by at least one processor, cause the at least one processor to perform:
- extracting a set of features from a table;
- applying a feature extraction model to the set of features to generate a pseudo sentence;
- applying a classifier model to the pseudo sentence to generate a label prediction for the table, the classifier model being trained with an augmented training pseudo sentence, the classifier model comprising a neural network model comprising a plurality of layers, the plurality of layers including a plurality of vectorization layers comprising at least one of:
- a language model comprising at least an embedding layer and a transformer layer or an encoder comprising at least one or more hidden layers, to process the pseudo sentence and generate at least one classification vector, the classifier model being trained with the augmented training pseudo sentence created by:
- extracting a set of training features from a training table;
- sampling the set of training features to create a set of sampled features;
- creating a set of augmented features by:
- shuffling one or more of the set of training features or the set of sampled features;
- and
- dropping at least one of the one or more of the set of training features or the set of sampled features, the dropping being performed using weights identifying a likelihood that a feature is dropped;
- and
- applying the feature extraction model to the set of augmented features to create the augmented training pseudo sentence;
- and
- updating a label of the table with the label prediction.
11 more elements
Dependencies are inferred from the claim text.