Systems and methods for detection of phishing webpages using machine learning
Disclosed herein are systems and method for reference-based detection of phishing webpages. In one aspect, a method includes inputting, for a webpage with a first domain, (1) textual data into a machine learning model (MLM) that outputs a first vector with probabilities of the textual data being associated with known brands, (2) HTML data into an MLM that outputs a second vector with probabilities of the HTML data being associated with the known brands, (3) at least one image into an MLM that outputs a third vector with probabilities of the at least one image being associated with the known brands. The model may input the first, second, and third vectors into an MLM that outputs a brand of the webpage. The method may block the webpage in response to determining that the first domain of the webpage does not match at least one domain corresponding to the brand.
1 . A method for detection of phishing webpages using machine learning, the method comprising:
extracting textual data, HTML data, and at least one image of a webpage with a first domain;
inputting the textual data into a first machine learning model configured to output a first vector comprising, for each respective brand of a plurality of brands, a respective probability that the textual data is associated with the respective brand;
inputting the HTML data into a second machine learning model configured to output a second vector comprising, for each respective brand of the plurality of brands, a respective probability that the HTML data is associated with the respective brand;
ranking each of the at least one image based on a probability that a given image is a logo, wherein the at least one image comprises at least a first image and a second image, and wherein the first image has a higher probability and is ranked higher than the second image;
inputting the first image into a third machine learning model configured to output a third vector comprising, for each respective brand of the plurality of brands, a respective probability that the first image is associated with the respective brand, wherein the second image is not input into the third machine learning model in response to determining that at least one probability associated with the first image is greater than a threshold probability;
inputting the first vector, the second vector, and the third vector into a fourth machine learning model configured to output a brand that the webpage is associated with;
identifying at least one domain corresponding to the brand; and
blocking the webpage in response to determining that the first domain of the webpage does not match the at least one domain corresponding to the brand.
2 . The method of claim 1 , wherein ranking each of the at least one image further comprises:
inputting the at least one image into a fifth machine learning model configured to output a fourth vector comprising, for each respective image of the at least one image, a respective probability that the respective image is the logo;
ranking each of the at least one image based on the fourth vector, wherein images with a higher probability of being a logo are ranked higher than other images.
3 . The method of claim 1 , further comprising:
in response to determining that no probability associated with the first image is greater than the threshold probability, inputting the second image into the third machine learning model.
4 . The method of claim 1 , further comprising:
in response to determining that the first domain of the webpage does not match the at least one domain corresponding to the brand, identifying the webpage as a phishing webpage;
adding an identifier of the webpage in a block list of webpages to automatically block.
5 . The method of claim 1 , further comprising:
in response to determining that the first domain of the webpage matches the at least one domain corresponding to the brand, generating the webpage for display.
6 . The method of claim 1 , further comprising:
in response to determining that the first domain of the webpage matches the at least one domain corresponding to the brand, identifying the webpage as a safe webpage; and
adding an identifier of the webpage in a white list of webpages to not block.
7 . The method of claim 1 , wherein extracting the textual data, the HTML data, and the at least one image of the webpage is in response to receiving a request to generate the webpage for display.
8 . A system for detection of phishing webpages using machine learning, comprising:
at least one memory;
at least one hardware processor coupled with the at least one memory and configured, individually or in combination, to:
extracting textual data, HTML data, and at least one image of a webpage with a first domain;
input the textual data into a first machine learning model configured to output a first vector comprising, for each respective brand of a plurality of brands, a respective probability that the textual data is associated with the respective brand;
input the HTML data into a second machine learning model configured to output a second vector comprising, for each respective brand of the plurality of brands, a respective probability that the HTML data being is associated with the respective brand;
rank each of the at least one image based on a probability that a given image is a logo, wherein the at least one image comprises at least a first image and a second image, and wherein the first image has a higher probability and is ranked higher than the second image;
inputting the first image into a third machine learning model configured to output a third vector comprising, for each respective brand of the plurality of brands, a respective probability that the first image is associated with the respective brand, wherein the second image is not input into the third machine learning model in response to determining that at least one probability associated with the first image is greater than a threshold probability;
input the first vector, the second vector, and the third vector into a fourth machine learning model configured to output a brand that the webpage is associated with;
identify at least one domain corresponding to the brand; and
block the webpage in response to determining that the first domain of the webpage does not match the at least one domain corresponding to the brand.
9 . The system of claim 8 , wherein ranking each of the at least one image further comprises:
inputting the at least one image into a fifth machine learning model configured to output a fourth vector comprising, for each respective image of the at least one image, a respective probability that the respective image is the logo;
ranking each of the at least one image based on the fourth vector, wherein images with a higher probability of being a logo are ranked higher than other images.
10 . The system of claim 8 , further comprising:
in response to determining that no probability associated with the first image is greater than the threshold probability, inputting the second image into the third machine learning model.
11 . The system of claim 8 , further comprising:
in response to determining that the first domain of the webpage does not match the at least one domain corresponding to the brand, identifying the webpage as a phishing webpage;
adding an identifier of the webpage in a block list of webpages to automatically block.
12 . The system of claim 8 , further comprising:
in response to determining that the first domain of the webpage matches the at least one domain corresponding to the brand, generating the webpage for display.
13 . The system of claim 8 , further comprising:
in response to determining that the first domain of the webpage matches the at least one domain corresponding to the brand, identifying the webpage as a safe webpage; and
adding an identifier of the webpage in a white list of webpages to not block.
14 . The system of claim 8 , wherein extracting the textual data, the HTML data, and the at least one image of the webpage is in response to receiving a request to generate the webpage for display.
15 . A non-transitory computer readable medium storing thereon computer executable instructions for detection of phishing webpages using machine learning, including instructions for:
extracting textual data, HTML data, and at least one image of a webpage with a first domain;
inputting the textual data into a first machine learning model configured to output a first vector comprising, for each respective brand of a plurality of brands, a respective probability that the textual data is associated with the respective brand;
inputting the HTML data into a second machine learning model configured to output a second vector comprising, for each respective brand of the plurality of brands, a respective probability that the HTML data is associated with the respective brand;
ranking each of the at least one image based on a probability that a given image is a logo, wherein the at least one image comprises at least a first image and a second image, and wherein the first image has a higher probability and is ranked higher than the second image;
inputting the first image into a third machine learning model configured to output a third vector comprising, for each respective brand of the plurality of brands, a respective probability that the first image is associated with the respective brand, wherein the second image is not input into the third machine learning model in response to determining that at least one probability associated with the first image is greater than a threshold probability;
inputting the first vector, the second vector, and the third vector into a fourth machine learning model configured to output a brand that the webpage is associated with;
identifying at least one domain corresponding to the brand; and
blocking the webpage in response to determining that the first domain of the webpage does not match the at least one domain corresponding to the brand.