Large margin perceptrons for document categorization
AssigneeMicrosoft Corporation
InventorsHugo Zaragoza, Ralf Herbrich
Life of this patent
Estimating the term…
Abstract
A modified large margin perceptron learning algorithm (LMPLA) uses asymmetric margin variables for relevant training documents (i.e., referred to as “positive examples”) and non-relevant training documents (i.e., referred to as “negative examples”) to accommodate biased training sets. In addition, positive examples are initialized to force at least one update to the initial weighting vector. A noise parameter is also introduced to force convergence of the algorithm.
Claims24
A computer program product encoding a computer program for executing on a computer system a computer process for computing a classification weighting vector from a training set of training documents, the computer process comprising:
- determining whether one of the training documents introduces an error within two distinct margins relative to a inner product associated with the one of the training documents;
- and
- updating the classification weighting vector based on a document vector of the one of the training documents, if the one of the training documents introduces an error within the two distinct margins.
A method of computing a classification weighting vector from a training set of training documents, the method comprising:
- determining whether one of the training documents introduces an error within two distinct margins relative to a inner product associated with the one of the training documents;
- and
- updating the classification weighting vector based on a document vector of the one of the training documents, if the one of the training documents introduces an error within the two distinct margins.
A classification system for computing a category weighting vector associated with a given category from a training set of training documents, the system comprising:
- a training module generating document vectors from the training set and generating the category weighting vector based at least two asymmetric margins.
A computer program product encoding a computer program for executing on a computer system a computer process for computing a classification weighting vector from a training set of training documents, the computer process comprising:
- initializing the classification weighting vector based on one or more document vectors of relevant training documents from the training set;
- determining whether one of the training documents introduces an error within a margin relative to a perturbed inner product associated with the one of the training documents;
- and
- updating the classification weighting vector based on a document vector of the one of the training documents, if the one of the training documents introduces an error within the margin.
A method of computing a classification weighting vector from a training set of training documents, the method comprising:
- initializing the classification weighting vector based on one or more document vectors of relevant training documents from the training set;
- determining whether one of the training documents introduces an error within a margin relative to a perturbed inner product associated with the one of the training documents;
- and
- updating the classification weighting vector based on a document vector of the one of the training documents, if the one of the training documents introduces an error within the margin.
A classification system for computing a category weighting vector associated with a given category from a training set of training documents, the system comprising:
- a training module initializing the classification weighting vector based on one or more document vectors of relevant training documents from the training set, determining whether one of the training documents introduces an error within a margin relative to a perturbed inner product associated with the one of the training documents, and updating the classification weighting vector based on a document vector of the one of the training documents, if the one of the training documents introduces an error within the margin.
A computer program product encoding a computer program for executing on a computer system a computer process for computing a classification weighting vector from a training set of training documents, the computer process comprising:
- determining whether one of the training documents introduces an error within a margin relative to a inner product associated with the one of the training documents;
- and
- updating the classification weighting vector based on a document vector of the one of the training documents, if the one of the training documents introduces an error within the margin;
- and
- introducing a noise parameter into the inner product.
A method of computing a classification weighting vector from a training set of training documents, the method comprising:
- determining whether one of the training documents introduces an error within a margin relative to a inner product associated with the one of the training documents;
- and
- updating the classification weighting vector based on a document vector of the one of the training documents, if the one of the training documents introduces an error within the margin;
- and
- introducing a noise parameter into the inner product.
A classification system for computing a category weighting vector associated with a given category from a training set of training documents, the system comprising:
- a training module determining whether one of the training documents introduces an error within a margin relative to a perturbed inner product associated with the one of the training documents, updating the classification weighting vector based on a document vector of the one of the training documents, if the one of the training documents introduces an error within the margin, and introducing a noise parameter into the inner product.
Dependencies are inferred from the claim text.