Verifying and correcting training data for text classification

A non-transitory computer readable storage medium comprising a computer readable program for text classification, wherein the computer readable program when executed on a computer causes the computer to perform the steps of:

• identifying a text sample that has been derived from another text sample in a first text class of a plurality of hierarchical graphs of text classes, the identified text sample having a class label that identifies the first text class, wherein the identified text sample is more similar to a second text class in the plurality of hierarchical graphs of text classes than to the first text class, according to a similarity metric;
• deleting a word from the identified text sample that is associated with the second text class, to move the identified text sample away from the second text class in a semantic space, causing the identified text sample to more closely conform to a hierarchical graph of classes that includes the first class;
• adding a word to the identified text sample that is associated with the first text class, to move the identified text sample toward the first text class in the semantic space, causing the identified text sample to more closely conform to a hierarchical graph of classes that includes the first class;
• training a text classification model to assign class labels to samples, using the text sample as training data, to provide superior text classification performance in discriminating between the first class and the second class;
• and
• performing text classification on an input text using the trained text classification model implemented on a processor device.

A text classification system, comprising:

• a hardware processor;
• and
• a memory that stores a computer program product, which, when executed by the hardware processor, causes the hardware processor to:
• identify a text sample that has been derived from another text sample in a first text class of a plurality of hierarchical graphs of text classes, the identified text sample having a class label that identifies the first text class, wherein the identified text sample is more similar to a second text class in the plurality of hierarchical graphs of text classes than to the first class according to a similarity metric;
• delete a word from the identified text sample that is associated with the second class, which moves the identified text sample away from the second text class in a semantic space and causes the identified text sample to more closely conform to a hierarchical graph of classes that includes the first class,
• add a word to the identified text sample that is associated with the first class, which moves the identified text sample toward the first text class in the semantic space and causes the identified text sample to more closely conform to a hierarchical graph of classes that includes the first class;
• train a text classification model to assign labels to samples using the text sample as training data, to provide superior text classification performance in discriminating between the first class and the second class;
• and
• perform text classification on an input text using the trained text classification model implemented on a processor device.