Identification of fraudulent online profiles

A method, comprising:

• training, by a computer system included as part of a web-based service, a contrastive loss function to compare a submitted image to a plurality of images stored in a database, the contrastive loss function to be used to identify fraudulent images associated with a particular entity;
• receiving, by the computer system via a network connection, a request to create a new account profile, the request including new profile information with a new profile image and a new profile identifier;
• generating, by the computer system, a vector representation of the new profile image, wherein generating the vector representation of the new profile image includes using the contrastive loss function for:
• sending a plurality of pixel data of the new profile image to one or more convolution layers;
• generating corresponding outputs representative of characteristics of the plurality of pixel data;
• and
• reducing the corresponding outputs into one or more vectors indicative of the characteristics of the plurality of pixel data;
• determining, by the computer system, an image metric indicative of a degree of similarity between the vector representation of the new profile image and a vector representation of an existing profile image associated with a particular account;
• in response to determining that the determined image metric is not an exact match but satisfies a threshold value, determining, by the computer system, whether a difference between the new profile identifier and an existing profile identifier associated with the particular account satisfies a threshold identifier metric;
• and
• in response to determining that the threshold identifier metric is satisfied, determining, by the computer system, whether to reject the request to create the new account profile based on a degree of similarity of the new profile image to at least one of the one or more existing profile images.

A non-transitory, computer-readable medium having instructions stored thereon that are executable by a computer system included as part of a web-based service to perform operations comprising:

• training a contrastive loss function to identify images from a first entity that are similar to one or more images, stored in a database, that are associated with a second entity;
• receiving, via a network connection, a request to create a new account profile, the request including new profile information that includes a new profile image and a new profile identifier;
• comparing the new profile identifier to one or more existing profile identifiers;
• in response to determining that the new profile identifier is not an exact match but satisfies a threshold level of similarity to a particular existing profile identifier associated with a particular account, generating a vector representation of the new profile image, wherein generating the vector representation of the new profile image includes using the contrastive loss function for:
• generating respective outputs representative of different characteristics of a plurality of pixel data of the new profile image;
• and
• reducing a number of the outputs by combining the outputs into one or more vectors that are indicative of the different characteristics;
• comparing the vector representation of the new profile image to vector representations of one or more existing profile images associated with the particular account;
• generating a fake metric value that is indicative of a degree of similarity of the new profile image to the one or more existing profile images;
• and
• in response to determining that the fake metric value satisfies a threshold level of similarity to a particular existing profile image associated with the particular account, rejecting the request to create the new account profile.