Multi-feature balancing for natural language processors
Techniques are disclosed for systems including techniques for multi-feature balancing for natural langue processors. In an embodiment, a method includes receiving a natural language query to be processed by a machine learning model, the machine learning model utilizing a dataset of natural language phrases for processing natural language queries, determining, based on the machine learning model and the natural language query, a feature dropout value, generating, and based on the natural language query, one or more contextual features and one or more expressional features that may be input to the machine learning model, modifying at least one or the one or more contextual features and the one or more expressional features based on the feature dropout value to generate a set of input features for the machine learning model, and processing the set of input features to cause generating an output dataset for corresponding to the natural language query.
1. A computer-implemented method, comprising:
receiving a natural language query to be processed by a machine learning model, the machine learning model utilizing a dataset of natural language phrases for processing natural language queries;
determining, based on the natural language query, a feature dropout value for the machine learning model;
generating, based on the natural language query, a set of contextual features comprising a plurality of contextual features and a set of expressional features comprising one or more expressional features;
determining a feature correspondence between the plurality of contextual features and the one or more expressional features, to form a subset of contextual features comprising one or more contextual features corresponding to the one or more expressional features with which the feature correspondence is determined, among the plurality of contextual features;
removing, from the set of contextual features, at least a portion of the one or more contextual features of the subset of contextual features at a rate corresponding to the feature dropout value, to generate a set of modified contextual features;
generating a set of input features to include the set of modified contextual features and the one or more expressional features;
inputting, to the machine learning model, the set of input features; and
processing, by the machine learning model, the set of input features, to generate an output with respect to the natural language query,
wherein the feature dropout value comprises a first contextual feature dropout value corresponding to a first percentage of the plurality of contextual features of the set of contextual features.
2. The computer-implemented method of claim 1 , wherein:
the feature dropout value further comprises a second contextual feature dropout value corresponding to a second percentage of the plurality of contextual features of the set of contextual features, and
the removing further comprises removing the second percentage of the plurality of contextual features from the set of modified contextual features, to generate the set of modified contextual features.
3. The computer-implemented method of claim 1 , wherein:
the feature dropout value further comprises a first expressional feature dropout value corresponding to a percentage of expressional features of the one or more expressional features;
the computer-implemented method further comprises modifying the set of expressional features by removing the percentage of expressional features from the set of expressional features, to generate a set of modified expressional features; and
the set of input features is generated from the set of modified contextual features and the set of modified expressional features.
4. The computer-implemented method of claim 1 , further comprising:
comparing the dataset of natural language phrases to a training dataset used to train the machine learning model; and
determining, based on the comparison, a noise value corresponding to a number of natural language phrases in the dataset of natural language phrases and the training dataset associated with the same particular category and a number of natural language phrases in the dataset of natural language phrases and the training dataset associated with different categories,
wherein the feature dropout value is determined at least in part based on the noise value.
5. The computer-implemented method of claim 1 , wherein the machine learning model is a convolution neural network machine learning model and the set of input features corresponds to input nodes of a convolutional neural network.
6. A non-transitory computer-readable memory storing a plurality of instructions executable by one or more processors, the plurality of instructions comprising instructions that, when executed by the one or more processors, cause the one or more processors to perform a method including:
receiving a natural language query to be processed by a machine learning model, the machine learning model utilizing a dataset of natural language phrases for processing natural language queries;
determining, based on the natural language query, a feature dropout value for the machine learning model;
generating, based on the natural language query, a set of contextual features comprising a plurality of contextual features and a set of expressional features comprising one or more expressional features;
determining a feature correspondence between the plurality of contextual features and the one or more expressional features, to form a subset of contextual features comprising one or more contextual features corresponding to the one or more expressional features with which the feature correspondence is determined, among the plurality of contextual features;
removing, from the set of contextual features, at least a portion of the one or more contextual features of the subset of contextual features at a rate corresponding to the feature dropout value, to generate a set of modified contextual features;
generating a set of input features to include the set of modified contextual features and the one or more expressional features;
inputting, to the machine learning model, the set of input features; and
processing, by the machine learning model, the set of input features, to generate an output with respect to the natural language query,
wherein the feature dropout value comprises a first contextual feature dropout value corresponding to a first percentage of the plurality of contextual features of the set of contextual features.
7. The non-transitory computer-readable memory of claim 6 , wherein:
the feature dropout value further comprises a second contextual feature dropout value corresponding to a second percentage of the plurality of contextual features of the set of contextual features, and
the removing further comprises removing the second percentage of the plurality of contextual features from the set of modified contextual features, to generate the set of modified contextual features.
8. The non-transitory computer-readable memory of claim 6 , wherein:
the feature dropout value further comprises a first expressional feature dropout value corresponding to a percentage of expressional features of the one or more expressional features;
the method further includes modifying the set of expressional features by removing the percentage of expressional features from the set of expressional features, to generate a set of modified expressional features; and
the set of input features is generated from the set of modified contextual features and the set of modified expressional features.
9. The non-transitory computer-readable memory of claim 6 , wherein the method further includes:
comparing the dataset of natural language phrases to a training dataset used to train the machine learning model; and
determining, based on the comparison, a noise value corresponding to a number of natural language phrases in the dataset of natural language phrases and the training dataset associated with the same particular category and a number of natural language phrases in the dataset of natural language phrases and the training dataset associated with different categories,
wherein the feature dropout value is determined at least in part based on the noise value.
10. The non-transitory computer-readable memory of claim 6 , wherein the machine learning model is a convolution neural network machine learning model and the set of input features corresponds to input nodes of a convolutional neural network.