Learning form-based information classification
AssigneeIntuit Inc.
InventorsHoma Foroughi, Joy Rimchala
Life of this patent
Estimating the term…
Abstract
A method may include extracting training tokens from a training corpus, generating training token features and training contextual features for each of the training tokens, training a probabilistic graphical model using classes, the training token features, and the training contextual features, extracting, from actual data, actual tokens including an actual token, generating actual token features and actual contextual features for each of the actual tokens, and classifying the actual token as a class using the probabilistic graphical model, the actual token features, and the actual contextual features.
Claims20
A method, comprising:
- extracting, from actual data, a plurality of actual tokens comprising a first actual token;
- generating actual token features and actual contextual features for each of the plurality of actual tokens;
- and
- classifying the first actual token as a first class of a plurality of classes using the actual token features, the actual contextual features, and a probabilistic graphical model that was trained by:
- extracting training tokens from a training corpus;
- generating training token features and training contextual features for each of the training tokens;
- generating a first feature function comprising a first class predicate for determining whether a token in a sequence of tokens has the first class.
A system, comprising:
- a computer processor;
- a repository storing a training corpus, actual data, and a plurality of classes comprising a first class;
- a token extractor executing on the computer processor configured to:
- extract training tokens from the training corpus;
- and
- extract, from the actual data, a plurality of actual tokens comprising a first actual token;
- a feature generator executing on the computer processor configured to:
- generate training token features and training contextual features for each of the training tokens;
- and
- generate actual token features and actual contextual features for each of the plurality of actual tokens;
- and
- a classifier comprising a probabilistic graphical model and executing on the computer processor, wherein the classifier is configured to:
- train the probabilistic graphical model using the plurality of classes, the training token features, and the training contextual features, wherein training the probabilistic graphical model comprises:
- generating a first feature function comprising a first class predicate for determining whether a token in a sequence of tokens has a first one of a plurality of class types;
- and
- classify the first actual token as the first class using the probabilistic graphical model, the actual token features, and the actual contextual features.
A non-transitory computer readable medium comprising instructions that, when executed by a processor, perform:
- extracting training tokens from a training corpus;
- generating training token features and training contextual features for each of the training tokens;
- training a probabilistic graphical model using a plurality of classes, the training token features, and the training contextual features, wherein the training comprises:
- generating a first feature function comprising a first class predicate for determining whether a token in a sequence of tokens has a first one of a plurality of class types;
- extracting, from actual data, a plurality of actual tokens comprising a first actual token;
- generating actual token features and actual contextual features for each of the plurality of actual tokens;
- and
- classifying the first actual token as a first class of the plurality of classes using the probabilistic graphical model, the actual token features, and the actual contextual features.
Dependencies are inferred from the claim text.