Machine-Learning-Aided Automatic Taxonomy for Web Data

A system comprising:

• at least one hardware processor;
• and
• one or more software modules that are configured to, when executed by the at least one hardware processor,
• during a training mode, train a model to predict a class, from a plurality of classes in a taxonomy of web-based activities, based on a training dataset that comprises a plurality of annotated features, wherein each of the plurality of annotated features comprises one or more features, which have been derived from one or both of a uniform resource locator (URL) of an online resource or metadata associated with that online resource, and a ground-truth class assigned to those one or more features, and,
• during an operation mode,
• acquire web data comprising one or more activity records, wherein each of the one or more activity records comprises one or both of a URL of an online resource that was accessed by a visitor, or metadata associated with that online resource, and,
• for each of the one or more activity records,
• extract a set of one or more features from one or both of the URL or the metadata in the activity record,
• apply the trained model to the set of one or more features to predict a class, from the plurality of classes in the taxonomy, that is associated with the set of one or more features, and
• store the predicted class in association with the URL in the activity record.

A non-transitory computer-readable medium having instructions stored therein, wherein the instructions, when executed by a processor, cause the processor to:

• during a training mode, train a model to predict a class, from a plurality of classes in a taxonomy of web-based activities, based on a training dataset that comprises a plurality of annotated features, wherein each of the plurality of annotated features comprises one or more features, which have been derived from one or both of a uniform resource locator (URL) of an online resource or metadata associated with that online resource, and a ground-truth class assigned to those one or more features;
• and,
• during an operation mode,
• acquire web data comprising one or more activity records, wherein each of the one or more activity records comprises one or both of a URL of an online resource that was accessed by a visitor, or metadata associated with that online resource, and,
• for each of the one or more activity records,
• extract a set of one or more features from one or both of the URL or the metadata in the activity record,
• apply the trained model to the set of one or more features to predict a class, from the plurality of classes in the taxonomy, that is associated with the set of one or more features, and
• store the predicted class in association with the URL in the activity record.