Systems and methods for detecting fraudulent users
Inventors
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
Systems and methods provide for receiving data relating to an entity, wherein the data is categorized according to a plurality of fields, wherein each of the plurality of fields associated with the entity is defined by one of a plurality of field types, and where each of the fields are respectively associated with a plurality of field values. Field values are matched to a matching node in a graph database, wherein the graph database is defined by a plurality of nodes having one of the plurality of field types. The graph database is searched for fuzzy matches and based on finding one or more fuzzy matches, a fraud response action is executed.
Core Innovation
The invention retrieves account information of a plurality of users from a datastore and generates a graph from the retrieved account information. The graph includes a plurality of user nodes representing the plurality of users and a plurality of field nodes representing fields associated with the plurality of users, where each field node includes a value for its respective field. The method generates one or more edges between each user node and each field node associated with that user node.
For identifying cross-user field relationships, the method identifies a node pair in which a first field node is connected to a first user node and a second field node is connected to a second user node. The first field node and the second field node represent the same type of field. The method determines a similarity score between the first field node and the second field node for the node pair and connects the first field node and the second field node via an edge when the similarity score is more than a pre-specified threshold.
The method then determines one or more cliques in the graph based on an edge limit defining at least one clique. In related aspects, a computer-implemented method receives data relating to an entity categorized according to a plurality of fields, where each field is defined by a plurality of field types and each field is associated with a plurality of field values. The method matches field values to nodes in a graph database by searching for fuzzy matches among nodes having the same field type, and executes a fraud response action based on finding one or more fuzzy matches.
Claims Coverage
Two independent claims are identified (clm-00001 and clm-00008). Across these claims, the inventive features focus on graph construction with typed nodes and similarity-based edge creation, clique determination, fuzzy matching on a typed graph database, and executing fraud response actions based on matches.
Graph generation with typed user and field nodes and user–field edges
Generating a graph using the retrieved account information, including generating user nodes representing users, generating field nodes representing fields associated with the users where each field node includes a value for a respective field, and generating one or more edges between each user node and each field node associated with that user node.
Same-field-type similarity scoring with thresholded similarity edges
Identifying a node pair comprising a first field node connected to a first user node and a second field node connected to a second user node, where the first field node and the second field node represent the same type of field; determining a similarity score between the first field node and the second field node; and connecting the first field node and the second field node via an edge if the similarity score is more than a pre-specified threshold.
Clique determination based on an edge limit
Determining one or more cliques in the graph based on an edge limit defining at least one clique.
Fuzzy matching in a typed graph database and executing fraud response actions
Receiving data relating to an entity, categorizing the data according to a plurality of fields defined by a plurality of field types and associated with field values; matching at least one field value to a value associated with a matching node in a graph database with nodes defined by the plurality of field types where the matching node has the same type as the field associated with the field value; searching the graph database for fuzzy matches for each of the plurality of field values by comparing each field value to a subset of nodes having one of the plurality of field types matching the field type; and executing a fraud response action based on finding one or more fuzzy matches.
The independent claims collectively cover graph construction with user nodes and field nodes including field values, similarity edges between same field types when a similarity score exceeds a threshold, clique determination using an edge limit, and typed-field fuzzy matching in a graph database followed by fraud response actions when fuzzy matches are found.
Stated Advantages
Scalability advantages over exhaustive watchlist pairwise comparisons.
Documented Applications
Fraud detection using a fraud detection system that builds a graph from user/account data and triggers fraud response actions such as requesting human review or automated actions.
Reporting, where generated fraud response information is provided for review by human evaluators and can be associated with fraud response reports.
Account-level fraud responses, including account locking/freezing/closing and blocking transactions [as described in the provided summary].
Interested in licensing this patent?