Website categorization through semantic embeddings of text context

1 . A computer-implemented method in a web page security server on a data communication network, at least partially implemented in hardware, for securing web pages based on semantic embedding of text content, the method comprising:

crawling Universal Resource Locator (URL) pages of a web page host, of known categories on the Internet to obtain text web contents;

generating semantic embeddings by encoding the text web contents with semantic meaning, wherein the text web contents are transformed to vectors of numbers;

training a web page security model using the semantic embeddings, including: assigning the known categories; clustering the semantic embeddings around the known categories; and computing an average vector of all embeddings for each known category;

receiving an unknown web page;

generating semantic embeddings by enriching the unknown web page with semantic meaning in order to determine an unknown web page vector;

calculating proximity of the unknown web page semantic embeddings to known web categories by comparing the average vector against the unknown web page vector;

outputting a category label for a web site category for the new web page based on the calculated proximity, and

automatically applying rules of a security policy to block the unknown web page, based on the category label.

2 . The method of claim 1 , further comprising applying a filtering policy to the unknown web page based on the category label.

3 . The method of claim 1 , wherein outputting a category label comprises outputting highest ranked categories.

4 . A non-transitory computer-readable medium in a web page categorization security server, at least partially implemented in hardware and on a data communication network, at least partially implemented in hardware, storing code that when executed, performs a method for securing web pages through semantic embedding of text content, the method comprising:

crawling Universal Resource Locator (URL) pages of a web page host, of known categories on the Internet to obtain text web contents;

generating semantic embeddings by encoding the text web contents with semantic meaning, wherein the text web contents are transformed to vectors of numbers;

training a web page security model using the semantic embeddings, including:

assigning the known categories;

clustering the semantic embeddings around the known categories; and

computing an average vector of all embeddings for each known category;

receiving an unknown web page;

generating semantic embeddings by enriching the unknown web page with semantic meaning in order to determine an unknown web page vector;

calculating proximity of the unknown web page semantic embeddings to the known web categories by comparing the average vector against the unknown web page vector;

outputting a category label for a web site category for the new web page based on the calculated proximity, and

automatically applying rules of a security policy to block the unknown web page, based on the category label.

5 . A web page security server, on a data communication network, at least partially implemented in hardware, for securing web pages through semantic embedding of text content, the web page security server comprising:

a processor;

a network interface communicatively coupled to the processor and to a data communication network; and

a memory, communicatively coupled to the processor and storing:

a web crawling module to crawl Universal Resource Locator (URL) pages of a web page host, of known categories on the Internet to obtain text web contents;

a vector transformation module to generate semantic embeddings by encoding the text web contents of known categories with semantic meaning, wherein the text web contents are transformed to vectors of numbers;

a training module to train a web page security model using the semantic embeddings, including:

assigning the known categories;

clustering the semantic embeddings around the known categories; and

computing average vector of all embeddings for each known category;

a queue to receive an unknown web page,

wherein the vector transformation module generates semantic embeddings by enriching the unknown web page with semantic meaning in order to determine an unknown web page vector;

a distance module to calculate proximity of the unknown web page semantic embeddings to the known web categories by comparing the average vector against the unknown web page vector;

a category labeling module to output a category label for a web site category for the new web page based on the calculated proximity, and

a security policy module to automatically apply rules of a security policy to block the unknown web page, based on the category label.