Fraudulent page detection
AssigneeSentryBay Limited
InventorsDavid Lynch Waterson, Bevan J Collins, Raghuram Vijapurapu, Marcus Andrew Whittington
Life of this patent
Estimating the term…
Abstract
A method of determining whether a page is a fraudulent page comprising the steps of: extracting a plurality of tokens from the page, (403) for each token, calculating a token probability being the probability of the token being in a fraudulent page (404), using the calculated token probabilities, calculating a page probability being the probability of the page being a fraudulent page (405), wherein the token probability of a token being in a fraudulent page is calculated based on a number of fraudulent pages and a number of non-fraudulent pages which contain the token from a training corpus of fraudulent pages and non-fraudulent pages.
Claims10
A method of determining whether a page is a fraudulent page mimicking a page of a known genuine website, comprising the steps of:
- on a computer:
- receiving at least one token database corresponding to a genuine website, of an entity wanting to protect customers from fraudulent websites that mimic the genuine website, the token database being custom generated in respect of the genuine website by or on behalf of the entity to contain data relating to tokens that occur in pages of the genuine website and/or likely to occur in fraudulent pages of one or more fraudulent websites mimicking the genuine website,
- obtaining a page from a website purporting to be the genuine website, extracting a plurality of tokens from the page, for each token, determining a token probability from data in the token database corresponding to the genuine website, the token probability being a probability indicative of the token being in a page of the genuine website,
- wherein the token probability of a token being in a page from the genuine website is based on a number of fraudulent pages mimicking pages of the genuine website and a number of non-fraudulent pages corresponding to the genuine website and/or other genuine websites that contain the token, said fraudulent pages and non-fraudulent pages being in a training corpus provided by or on behalf of the entity, and
- wherein using token probabilities determined for each token, calculating a page probability indicating the similarity of the page to a non-fraudulent page corresponding to the genuine website, and
- if the page probability achieves or exceeds a similarity threshold and the URL of the website is not that of the genuine website, determining that the page is a fraudulent page mimicking a page of the genuine website.
A system for determining whether a page is a fraudulent page mimicking a page of a known genuine website, comprising an end user personal computer (PC), server or service adapted to download a page and adapted to:
- at a computer processor,
- receive at least one token database corresponding to a genuine website of an entity wanting to protect customers from fraudulent websites that mimic the genuine website, the token database being custom generated in respect of the genuine website by or on behalf of the entity to contain data relating to tokens that occur in pages of the genuine website and/or likely to occur in fraudulent pages of one or more fraudulent websites,
- obtaining a page from a website purporting to be the genuine website,
- extract a plurality of tokens from the page,
- for each token, determine a token probability from data in the token database, the token probability being a probability indicative of the token being in a page of the genuine website,
- wherein the token probability of a token being in a page from the genuine website is based on a number of fraudulent pages mimicking pages of the genuine website and a number of non-fraudulent pages corresponding to the genuine website and/or other genuine websites that contain the token, said fraudulent pages and non-fraudulent pages being in from a training corpus provided by or on behalf of the entity, and
- wherein using token probabilities determined for each token, calculating a page probability indicating the similarity of the page to a non-fraudulent page corresponding to the genuine website, and
- if the page probability achieves or exceeds a similarity threshold and the URL of the website is not that of the genuine website, determining that the page is a fraudulent page mimicking a page of the genuine website.
Dependencies are inferred from the claim text.