Back to results
US 11,887,061 B1Granted
Systems and methods for determination, description, and use of feature sets for machine learning classification systems, including electronic messaging systems employing machine learning classification
AssigneeZix Corporation
InventorPeter Gallagher McNeil
Life of this patent
Estimating the term…
Abstract
Systems and methods for dynamically defining features for machine learning models in classification systems are disclosed. Such systems and methods may be usefully applied to train and utilize machine learning models in a variety of contexts, including email classification systems.
Claims15
A method of feature set extraction, comprising:
- receiving a corpus of training content;
- parsing at least a portion of the training content to generate a plurality of features and a plurality of cross-tabulations for the features, wherein the training content comprises a plurality of training emails and the plurality of features comprises email addresses of the plurality of training emails, the email addresses comprising an email from address, an email to address, an email reply to address, an email cc address and an email bcc address, wherein the email addresses are divided into segments comprising a domain segment, a local segment, and a friendly segment;
- generating a value for each of the plurality of features by generating a unique training hash for the domain segment, a unique training hash for the local segment, and a unique training hash for the friendly segment of each of the email addresses in the plurality of training emails, each unique training hash associated with a feature value and a feature structure;
- generating a classification value for each of the cross-tabulations of the features;
- receiving a corpus of live content;
- parsing the live content based on the plurality of features to identify one or more of the features in the live content;
- generating a value for each of the identified one or more features in the live content;
- generating a classification value for each of the cross-tabulations of the identified one or more features in the live content;
- and
- based on the generated classification value for each of the cross-tabulations of the identified one or more features in the live content, classifying the live content.
7 more elements
A system for feature extraction, comprising:
- a processor;
- a non-transitory computer-readable medium;
- and
- stored instructions executable by the processor for executing:
- receiving a corpus of training content;
- parsing at least a portion of the training content to generate a plurality of features and a plurality of cross-tabulations for the features, wherein the training content comprises a plurality of training emails and the plurality of features comprises email addresses of the plurality of training emails, the email addresses comprising an email from address, an email to address, an email reply to address, an email cc address and an email bcc address, wherein the email addresses are divided into segments comprising a domain segment, a local segment, and a friendly segment;
- generating a value for each of the plurality of features by generating a unique training hash for the domain segment, a unique training hash for the local segment, and a unique training hash for the friendly segment of each of the email addresses in the plurality of training emails, each unique training hash associated with a feature value and a feature structure;
- generating a classification value for each of the cross-tabulations of the features;
- receiving a corpus of live content;
- parsing the live content based on the plurality of features to identify one or more of the features in the live content;
- generating a value for each of the identified one or more features in the live content;
- generating a classification value for each of the cross-tabulations of the identified one or more features in the live content;
- and
- based on the generated classification value for each of the cross-tabulations of the identified one or more features in the live content, classifying the live content.
11 more elements
A computer programming product for feature extraction comprising a non-transitory computer-readable medium storing instructions executable by a processor for:
- receiving a corpus of training content;
- parsing at least a portion of the training content to generate a plurality of features and a plurality of cross-tabulations for the features, wherein the training content comprises a plurality of training emails and the plurality of features comprises email addresses of the plurality of training emails, the email addresses comprising an email from address, an email to address, an email reply to address, an email cc address and an email bcc address, wherein the email addresses are divided into segments comprising a domain segment, a local segment, and a friendly segment;
- generating a value for each of the plurality of features by generating a unique training hash for the domain segment, a unique training hash for the local segment, and a unique training hash for the friendly segment of each of the email addresses in the plurality of training emails, each unique training hash associated with a feature value and a feature structure;
- generating a value for each of the plurality of features;
- generating a classification value for each of the cross-tabulations of the features;
- receiving a corpus of live content;
- parsing the live content based on the plurality of features to identify one or more of the features in the live content;
- generating a value for each of the identified one or more features in the live content;
- generating a classification value for each of the cross-tabulations of the identified one or more features in the live content;
- and
- based on the generated classification value for each of the cross-tabulations of the identified one or more features in the live content, classifying the live content.
8 more elements
Dependencies are inferred from the claim text.