Fraud prevention through disposable email address intelligence
The use of machine-learning models to analyze email attributes and generate risk scores for disposable email addresses addresses the risk of unauthorized access, enhancing security and stability in computing environments by improving access control.
Patent Information
- Application Number
- PCT/US2025/027799
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-05-05
- Publication Date
- 2026-11-12
AI Technical Summary
Disposable email addresses pose a risk for identity proofing services due to their potential for abuse by fraudulent actors, as they lack robust security measurements and can lead to unauthorized access and exposure of sensitive information, compromising user privacy.
A system and method using machine-learning models to analyze email attributes and generate risk scores for determining the likelihood of an email address being disposable, enabling improved access control in interactive computing environments.
Enhances security by accurately identifying disposable email addresses, reducing unauthorized access, and providing refined access controls, thereby improving the stability and security of computing environments.
Smart Images

Figure US2025027799_12112026_PF_FP_ABST
Abstract
Description
PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO FRAUD PREVENTION THROUGH DISPOSABLE EMAIL ADDRESS INTELLIGENCETechnical Field
[0001] The present disclosure relates generally to identity verification and fraud prevention. More specifically, but not by way of limitation, this disclosure relates to systems and methods for identity verification and fraud prevention through disposable email address intelligence and machine-learning.Background
[0002] Email addresses may be easily created and disposed of. Disposable email addresses are often referred to as temporary, throwaway, or burner emails. Such disposable emails may be used to create temporary accounts which lack identifying data. Disposable emails present risks towards identity proofing services. A primary concern with disposable emails is the potential for abuse by scammers, cybercriminals, and other fraudulent actors. The fraudulent actor can create fake identities or register fraudulent accounts on websites with malicious intent via a disposable email. The fraudulent actor can exploit the anonymity provided by the disposable email to engage in online fraud, phishing, identity theft, or other illicit activities. Disposable email services are also vulnerable to security' breaches. Because authentication services often lack robust security measurements directed to evaluating the authenticity of a given email address, unauthorized access to temporary email accounts is possible. Such access can lead to the exposure of sensitive information and compromise user privacy.Summary
[0003] Various embodiments of the present disclosure provide techniques for fraud prevention and improved computing system security through improved detection of disposable email addresses. A method for detecting disposable email addresses system can include receiving an input email address associated with an entity' and generating a set of email attributes based on the input email address. The method can include determining a risk score indicating a likelihood the input email address includes a disposable email address based on applying the set of email attributes to a machine-learning model trained on an email-based training data set, where the email-based training data set includes non-disposable 1US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO email addresses and disposable email addresses. The method can include transmitting a responsive message including at least the input email address and access information based on the risk score for use in controlling access of the entity to an interactive computing environment.
[0004] In other aspects, a system can include a processor and a non-transitory computer-readable medium that can include instructions that are executable by the processor to cause the processor to perform various operations. The system can receive an input email address associated with an entity and generate a set of email attributes based on the input email address. The computing system can determine a risk score indicating a likelihood the input email address includes a disposable email address based on applying the set of email attributes to a machine-learning model trained on an email-based training data set, where the email-based training data set includes non-disposable email addresses and disposable email addresses. The computing system can transmit a responsive message including at least the input email address and access information based on the risk score for use in controlling access of the entity7to an interactive computing environment.
[0005] In other aspects, a non-transitory7computer-readable medium can include instructions that are executable by a processing device for causing the processing device to perform various operations. The operations can include receiving an input email address associated with an entity and generating a set of email attributes based on the input email address. The operations can include determining a risk score indicating a likelihood the input email address includes a disposable email address based on applying the set of email attributes to a machine-learning model trained on an email-based training data set, where the email-based training data set includes non-disposable email addresses and disposable email addresses. The operations can include transmitting a responsive message including at least the input email address and access information based on the risk score for use in controlling access of the entity to an interactive computing environment.
[0006] This summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used in isolation to determine the scope of the claimed subject matter. The subject matter should be understood by reference to appropriate portions of the entire specification, any or all drawings, and each claim. The foregoing, together with other features and examples, will become more apparent upon referring to the following specification, claims, and accompanying drawings.2US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO Brief Description of the Drawings
[0007] FIG. 1 is a block diagram of an operating environment to implement a disposable email identification computing system, according to certain examples.
[0008] FIG. 2 is a flow diagram of a process for determining the authenticity of a user account via a trained machine-learning model, according to certain examples.
[0009] FIG. 3 provides an example of a plurality of components within a decomposed email address, according to certain examples.
[0010] FIG. 4 provides an example set of operations for training a machine-learning model to determine email disposability' risk scores, according to certain examples.
[0011] FIG. 5 is a block diagram depicting an example of a computing device, which can be used to implement operations performed by the disposable email identification computing system, according to certain examples.Detailed Description
[0012] Systems and methods for securing access in interactive computing environments via machine-learning based analysis of email addresses are described herein. The email address analysis can include identifying emails as disposable, which are email addresses more likely to be fraudulent, or non-disposable. which are email addresses more likely authentic and more permanently associated with the user providing the email address.Securing access in an interactive computing environment refers to authentication and access controls according to various techniques. Securing access can include denying access to disposable determined email accounts. Denying access can include blocking the creation of a user account associated with the disposable determined email address, restricting access to the IP address of the user associated with the disposable email address, and the like. Securing access can also include restricting, but not outright denying access to the disposable identified email address, such as by limiting access to certain features of the computing environment, or requiring additional authentication from the associated user account prior to granting greater access.
[0013] The interactive computing environment can refer to softw are-based applications, hardware-based systems, or a combination of software-based applications and hardwarebased systems. Software-based applications include those which a user may authenticate 3US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO themselves into, such as online video games, professional servicing applications, chat forums, and the like. Hardware computing environments can include servers, remote computing systems, and other devices communicatively coupled over a network. The interactive computing environments are generally accessed via authentication protocols based in part on user email addresses, or data pertaining to a user’s email address is otherwise accessible to the interactive computing environment. Therefore, interactive computing environments have access to the email addresses associated with the user, and according to the techniques described here, can control user access based on further analyses related to the email address.
[0014] The email address analyses described here apply various rules for decomposing email addresses into features, or analyzable portions which can be provided to one or more machine-learning models. The machine-learning models include software architecture which is able to score the validity of input email addresses based on previously provided examples (i. e. , machine-learning model training). Techniques for training the machine-learning models to generate email disposability scores are further described according to the various examples herein.
[0015] The scoring applied by the machine-learning model, used to control access to interactive computing environments, includes providing confidence scores as to whether an email address constitutes a non-disposable email address or a disposable email address. Nondisposable email addresses refer to those authentic to the user. Non-disposable email addresses are typically used to authenticate across a variety of services, and are more routinely accessed by the associated user compared to disposable email addresses. Further, non-disposable email addresses typically include more characteristics of personally identifiable information (“PII”) of the associated user, such as including the user’s first name or last name in the email address.
[0016] In contrast, disposable email addresses, also referred to as “burner” or “throwaway” email addresses, may be created and used to avoid giving out PII of the associated user. Disposable email addresses are also typically accessed and used less frequently compared to non-disposable emails. Because disposable emails typically do not include PII of the associated user, and because activity on disposable email addresses is limited, disposable emails are often linked to threat actors such as scammers and cybercriminals accessing interactive computing environments with fraudulent intent. Such users can exploit the anonymity offered by disposable email addresses while avoiding being 4US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO linked to the disposable email address. Disposable email addresses are themselves often prone to security breaches, further compounding the risks associated with disposable email addresses.
[0017] Properly controlling access to an interactive computing environment can improve the security of the interactive computing environment. For example, requiring a usemame / password combination, multi-factor authentication, biometric authentication, and the like can provide security for sensitive accounts or data included in the interactive computing environment. Frequently, the means for authenticating user accounts and associated entities is via the user account’s email address. Existing authentication techniques are premised on authenticating user accounts based on matching usernames and emails to provided passwords. These techniques assume that the email addresses are valid, and thus are limited in the range of methods in which user accounts can be validated. Therefore, interactive computing environments face exposure from disposable email addresses which may facially be valid, but in reality, may provide a vector for unauthorized access to the computing environment.
[0018] As a practical example, an interactive computing environment may include a chat forum or other environment allowing for communications between parties. The interactive computing environment may have policies regulating the permissible actions of user accounts, and may take action against user accounts violating the policies, such as by banning or otherwise denying access to the interactive computing environment. However, while a given user account may be denied access, the associated entity' which triggered the ban may then create a disposable email address and re-register into the same computing environment. Thus, bad actors can evade access denials within computing environments, and pass under electronic verification sen-ices via creation of disposable email accounts.
[0019] Certain aspects described herein for controlling access to the interactive computing environment by evaluating the disposability (i.e., authenticity) of email addresses can resolve one or more issues identified above. Various techniques using computing systems can be employed to output various information and controls responsive to received email addresses. For example, a computing device or system can predict email risk assessment values (e.g., risk scores associated with granting access to an online computing environment, or credit scores, etc.) with respect to one or more entities that can be used for making decisions about the entities. A risk assessment computing system can receive an input email 5US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO address associated with an entity. The input email address can be received as a specific inquiry, where a user may identify and submit a suspect email address for further risk evaluation. Additionally or alternatively, the email address authentication query can be received in tandem with a request to perform a digital interaction, such as a data access request, user authentication, transaction, or the like. The email address can be associated with an entity, where the entity may be a person or an organization acting officially via the email address and associated electronic persona. Email addresses associated with such entities may¬ be authentic (i.e., having a valid user profile corresponding to the email address) while alack of intent to use an email account as an electronic persona of the entity can indicate inauthenticity (i.e., where the email address was created to distance the email address from the entity, or where an associated entity does not exist outside of the email address).
[0020] The email address can then be processed by an email decomposition engine. The email decomposition engine includes programmable logic for receiving email addresses and decomposing received email addresses into a decomposed email data structure. The decomposed email data structure can include the received email addresses broken into separate components. In one example, the separate components can include a handler (e.g., username), sever name, domain, top level domain, and one or more labels. The separate components may overlap with one another. For instance, the domain can include the top level domain and the server name. The one or more labels can include the top level domain, decomposed via punctuation.
[0021] An attribute engine can then be applied to each decomposed email address to transform the decomposed email components into a plurality of features. The plurality of features may be used to characterize the email addresses from a variety of approaches. The plurality of features can include character analysis of various components and combinations of components of the email addresses. Further examples of feature engineering techniques, not present in existing systems, are described further below according to the examples provided in the detailed description.
[0022] The generated attributes for each email address can be applied to one or more machine-learning models trained on an email address training set, decomposed and analyzed according to the same described features. Aspects of the disclosure relate to particular techniques for training the machine-learning models(s) according to specific techniques to more accurately determine the likelihood an email address is a fraudulent, disposable email.6US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO For instance, training can include particular rules and conditions for down-sampling initial training data sets to improve the accuracy of the machine-learning models in generating risk score determinations. Such training techniques are described further below according to the examples provided in the detailed description.
[0023] The output risk score, associated with an input email address, can be used to improve the data security in various computing environments. In some examples, users with access to the disposable email identification computing system can provide input email addresses for disposability determinations, such that the user can interface in real-time with suspect email addresses and associated entities. In the same and other examples, the disposable email identification computing system can receive access requests from entities, where the requests in part include the email address for query ing. Then, determinations of whether to process the entity’s request may then be controlled based on the determined disposability of the email address. Access controls can include denying the associated email address access to the interactive computing environment, restricting access to the interactive computing environment, or flagging relevant software ow ners of risks associated with entities attempting to access the interactive computing environment.
[0024] Certain aspects described herein can improve the technical field of access control for a computing environment. For instance, by’ determining the disposability of a given email address tied to an entity, a risk assessment computing system may provide access to the interactive computing environment using fewer computing resources compared to other risk assessment systems. The described access control techniques provide for improved stability7and security within computing environments by providing an unconventional technique for assessing the authenticity of user profiles via associated email addresses. Compared to prior techniques, such an approach can provide another layer of security in refined access controls within the computing environment.
[0025] Additionally, specific techniques for training machine-learning models are described, providing technical improvements to machine-learning models trained for purposes of identity authentication. For one, the techniques described herein apply a unique set of features specifically tailored for assessing email disposability’ based on decomposed email addresses. Such feature engineering provides a first set of unique operations for improving machine-learning models to authenticate users. Moreover, aspects described herein include a multi-stage technique for training machine-learning models. In a first stage.7US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO machine-learning models can be trained using disposable and non-disposable emails. While, in a second stage, the training set is adjusted to include false positive detections (i. e. , emails incorrectly identified as disposable). Such multi-stage training can provide an enhanced user authentication machine-learning model that can detect disposable emails while limiting the number of false positives.
[0026] These illustrative examples are given to introduce the reader to the general subject matter discussed here and are not intended to limit the scope of the disclosed concepts. The following sections describe various additional features and examples with reference to the drawings in which like numerals indicate like elements, and directional descriptions are used to describe the illustrative examples but, like the illustrative examples, should not be used to limit the present disclosure.Operating Environment Example for Disposable Email Identification
[0027] Referring now to the drawings, FIG. 1 is a block diagram of an operating environment configured to implement a disposable email identification computing system, according to certain examples. In the operating environment 100, disposable email identification computing system 101 builds and trains models that can be used to determine whether an email address associated with an entity' is a disposable email, and further, fraud preventative actions to take in response to determined disposable emails. The disposable email identification computing system 101 can further apply one or more algorithms involving the several models to determine risk scores associated with email addresses indicating their disposability' (also referred to as disposable email risk scores).
[0028] FIG. 1 depicts examples of hardware components of the disposable email identification computing system 101. The disposable email identification computing system 101 is a specialized computing system that may be used for processing large amounts of data using a large number of computer processing cycles. The disposable email identification computing system 101 can include a model training server 108 for building and training machine-learning models 114 including classification models in addition to other machinelearning models. The disposable email identification computing system 101 can further include an email evaluation server 106 for assessing the likelihood of an input email address being a disposable email address, based on attributes associated with an input email address.8US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO
[0029] The model training server 108 can include one or more processing devices that execute program code, such as a model training application 112. The program code is stored on a non-transitory computer-readable medium. The model training application 112 can execute one or more processes to execute and / or retrain a machine-learning model 114 for assessing the validity of an input email address.
[0030] In some aspects, the model training application 112 can build and train a machinelearning model 114 utilizing model training samples (e.g., including disposable emails 120 and non-disposable emails 122). The model training samples can be stored in one or more network-attached storage units on which various repositories, databases, or other structures are stored. Examples of these data structures are the email address repository 118.Additionally model training samples including the reduced sets of attributes and a reduced training set 126 can be generated and fdtered according to various described techniques, such as those discussed with respect to FIG. 4.
[0031] Network-attached storage units may store a variety of different types of data organized in a variety of different ways and from a variety of different sources. For example, the network-attached storage unit may include storage other than primary' storage located within the model training server 108 that is directly accessible by processors located therein. In some aspects, the network-attached storage unit may include secondary, tertiary, or auxiliary storage, such as large hard drives, servers, virtual memory, among other types. Storage devices may include portable or non-portable storage devices, optical storage devices, and various other mediums capable of storing and containing data. A machine-readable storage medium or computer-readable storage medium may include a non-transitory medium in which data can be stored and that does not include carrier waves or transitory electronic signals. Examples of a non-transitory medium may include, for example, a magnetic disk or tape, optical storage media such as a compact disk or digital versatile disk, flash memory, memory, or memory devices.
[0032] The email evaluation server 106 can include one or more processing devices that execute program code, such as an email evaluation application 110. The program code is stored on a non-transitory computer-readable medium. The email evaluation application 110 can execute one or more processes to utilize the machine-learning model 114 trained by the model training application 112 based on disposable emails 120 and non-disposable emails 122, where the disposable emails 120 and non-disposable emails 122 can form the initial 9US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO training set 124, subsequently reduced by the model training application 112 in the reduced training set 126.
[0033] Furthermore, the disposable email identification computing sy stem 101 can communicate with various other computing systems, such as client computing systems 102. For example, client computing systems 102 may send authentication queries to email evaluation sen- er 106 to determine whether an email associated with an entity is disposable or may send signals to the email evaluation server 106 that control or otherwise influence different aspects of the disposable email identification computing system 101. The client computing systems 102 may also interact with user computing systems 130 via one or more public data networks 104 to facilitate interactions between users of the user computing systems 130 and interactive computing environments provided by the client computing systems 102.
[0034] Each client computing system 102 may include one or more third-party devices, such as individual servers or groups of servers operating in a distributed manner. A client computing system 102 can include any computing device or group of computing devices operated by a seller, lender, or other providers of products or services. The client computing system 102 can include one or more sen- er devices. The one or more server devices can include or can otherwise access one or more non-transitory computer-readable storage medium or media for storing executable program code. The client computing system 102 can also execute instructions that provide an interactive computing environment accessible to user computing systems 130. Examples of the interactive computing environment include a mobile application specific to a particular client computing system 102, a web-based application accessible via a mobile device, etc. The executable instructions are stored in one or more non-transitory computer-readable media.
[0035] The client computing system 102 can further include one or more processing devices that are capable of providing the interactive computing environment for performing operations described herein. The interactive computing environment can include executable instructions stored in one or more non-transitory computer-readable media. The instructions providing the interactive computing environment can configure one or more processing devices to perform operations described herein. In some aspects, the executable instructions for the interactive computing environment can include instructions that provide one or more graphical interfaces. The graphical interfaces are used by a user computing system 130 to 10US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO access various functions of the interactive computing environment. For instance, the interactive computing environment may transmit data to and receive data from a user computing system 130 to shift between different states of the interactive computing environment, where the different states allow one or more electronics transactions between the user computing system 130 and the client computing system 102 to be performed.
[0036] In some examples, a client computing system 102 may have other computing resources associated therewith (not shown in FIG. 1), such as server computers hosting and managing virtual machine instances for providing cloud computing services, server computers hosting and managing online storage resources for users, server computers for providing database services, and others. The interaction between the user computing system 130 and the client computing system 102 may be performed through graphical user interfaces presented by the client computing system 102 to the user computing system 130, or through an application programming interface (“APT’) calls or web service calls.
[0037] A user computing system 130 can include any computing device or other communication device operated by a user, such as a consumer or a customer. The user computing system 130 can include one or more computing devices, such as laptops, smartphones, and other personal computing devices. A user computing system 130 can include executable instructions stored in one or more non-transitory computer-readable media. The user computing system 130 can also include one or more processing devices that are capable of executing program code to perform operations described herein. In various examples, the user computing system 130 can allow a user to access certain online services from a client computing system 102 or other computing resources, to engage in mobile commerce with a client computing system 102, to obtain controlled access to electronic content hosted by the client computing system 102, etc.
[0038] In some examples, the disposable email identification computing system 101 can cause the user computing system 130, the client computing system 102, or a combination thereof to execute one or more actions in accordance with the disposable email risk score. For instance, the disposable email identification computing system 101 can selectively grant or deny access to various electronic functions based on the risk score. Based on the request associated with the email address, the disposable email identification computing system 101 can determine an entity profile for the entity that submitted the request via the user computing system 130 or the client computing system 102. The user computing system 130.11US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO the client computing system 102, or a combination thereof can then determine whether to grant the access request of the user computing system 130 to certain features of an interactive computing environment.
[0039] Each communication within the operating environment 100 may occur over one or more data networks, such as a public data network 104, an internal data network 116, or some combination thereof. A data network may include one or more of a variety of different ty pes of networks, including a wireless network, a wired network, or a combination of a wired and wireless network. Examples of suitable networks include the Internet, a personal area network, a local area network (“LAN”), a wide area network (“WAN”), or a wireless local area netw ork (“WLAN”). A wireless network may include a wireless interface or a combination of wireless interfaces. A wired network may include a wired interface. The wired or wireless networks may be implemented using routers, access points, bridges, gateways, or the like, to connect devices in the data network.
[0040] The number of devices depicted in FIG. 1 is provided for illustrative purposes. Different numbers of devices may be used. For example, while certain devices or systems are shown as single devices in FIG. 1, multiple devices may instead be used to implement these devices or systems. Similarly, devices or systems that are shown as separate, such as the email evaluation server 106 and model training server 108, may be instead implemented in a signal device or system.Example Operations for User Account Authentication Based on Email Disposability
[0041] FIG. 2 is a flow' diagram of a process for determining the authenticity of a user account via a trained machine-learning model, according to certain examples. For illustrative purposes, the process 200 is described with reference to implementations described above with respect to one or more examples described herein. Other implementations, however, are possible. In some aspects, the operations in FIG. 2 may be implemented in program code that is executed by one or more computing devices such as the email evaluation server 106 depicted in FIG. 1. In some aspects of the present disclosure, one or more operations shown in FIG. 2 may be omitted or performed in a different order. Similarly, additional operations not shown in FIG. 2 may be performed.
[0042] At block 202, the process 200 involves receiving an input email address associated with an entity. The input email address can be received as a specific inquiry,12US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO where a user may identify and submit a suspect email address for further risk evaluation to the user computing system 130 for investigation at the disposable email identification computing system 101, or can be received in tandem with a request to perform in a digital interaction, such as a data access request, user authentication, transaction, or the like. The email address can be associated with an entity, where the entity may be a person or an organization. The entity can be an authentic entity (i.e., having a valid user profile corresponding to the email address) or can be an inauthentic entity (i.e., where the email address was created to distance the email address from the inauthentic entity, or where the inauthentic entity does not exist outside of the email address).
[0043] At block 204, the process 200 involves decomposing the input email address into a decomposed email data structure including a plurality of components. Hardcoded logic within the email evaluation application 110 can be configured to generate different components of the plurality of components based on parsing the email address for specific characters. Separating email addresses into a first component including the handler from a second component including a domain can be performed based on parsing the input email address for the at sign (“@”). Top level domains can be identified and separated based on consulting a libraries of server names and top level domains. Additionally or alternatively, input email addresses can be stored and analyzed for identifying patterns within domains to identify server names and top level domain. A threshold quantity of phrases of character patterns can be compared against thresholds to identify new server names and top level domains for addition to corresponding libraries. FIG. 3 provides an example of a plurality of components within a decomposed email address, according to certain examples.
[0044] In the example of FIG. 3, a first email address “John.Doel23@test-fake.mail.com” is shown decomposed into multiple components. The components include the email handler ‘John.Doel23” and an email domain “test-fake.mail.com” demarcated by the at sign (“@”) from the remaining portion of the email address. The domain can be further decomposed into multiple components including a prefix “test-fake” and suffix “mail.com” and into a server name “test-fake, mail” and atop level domain “ com.” Labels may also be used to decompose the email address along other dimensions, where alphanumeric string can be divided by non-alphanumeric characters and punctuation. When applicable, the email evaluation application 110 can further decompose the handler into first name and last name components. As shown, components can overlap, such as where a domain component 13US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO includes a server name component and a top level domain component. The example components of FIG. 3 are illustrative, and it is to be appreciated that more or fewer components can form the plurality of components, including those shown in FIG. 3, in addition to other email address components.
[0045] Returning to FIG. 2, at block 206, the process 200 involves generating a set of email attributes based on the plurality of components. The set of email attributes can include attributes based entirely on a single component of the decomposed components produced per block 204, or a combination of components (e.g., an attribute corresponding to the handler and server name, but not the top level domain). In some examples, some components may be based on the entirety of the email address string such that decomposition per block 204 is an optional procedure. The sets of email attribute can include numerous ty pes of attributes, each for analyzing the email address string from a variety of perspectives. The following list is intended to provide example attributes that can be used to form the set of email attributes, though it is to be appreciated that each attribute in the list is neither necessary' nor sufficient to create the set of email attributes, and the set of email attributes can include more, or fewer attributes, including those not provided in the example list.
[0046] A first email attribute type can include string length attributes, measuring the number of all characters (alphanumeric or non-alphanumeric) evaluated in the domain name, handler, or both the domain name and handler. When measuring the string length of the domain name and handler, decomposition of the email address may not be necessary'. In some examples, a length ratio attribute can be generated by comparing the length of the handler to the domain name.
[0047] A second email attribute type can include depth attributes. The depth attribute can measure the number of elements separated by punctuation, or more specifically, periods or hyphens. Each component of the decomposed email address can have a separate depth attribute such that the email evaluation application 110 can generate attributes pertaining to domain depth, prefix depth, suffix depth, server name depth, handler depth, and the like.
[0048] A third email attribute type can include ratio attributes. The ratio attributes can measure the relative number of vowels, consonants, alphanumeric to non-alphanumeric characters, or other types of characters in each of the decomposed components. For instance, a ratio of vowels to consonants can be measured, ratios of letters to digits can similarly be 14US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO measured, and similar ratios determined. Moreover, the ratios relative to each component can be measured, such as the ratio of vowels to consonants in the handler compared to the domain. A second type of ratio attributes can include consecutive character-type attributes, measuring the string length of a consecutive character type in each component. For instance, a string of length greater than two digits can be compared against a string length of consecutive alphabetical characters. In the example of FIG. 3, the domain component has a ratio of 0 for numeric: alphabetical characters while the handler has a ratio of .273.
[0049] A fourth email attribute type can include entropy attributes, measuring the chaos of each component or the email address as a whole. Entropy can be used to evaluate the randomness, complexity, or uniqueness of each string component. Various entropy measuring techniques can be applied including calculating entropy via Shannon entropy, per character entropy, symbol distribution entropy, cross-entropy, and the like. In some examples, entropy can be determined by evaluating a given email address and its constituent components against a set of email addresses stored in the email address repository 118. For instance, crossentropy metrics can measure the divergence between the input email address and one or more email addresses stored in the email repository.
[0050] Fifth and sixth email attribute types can include distinct character attributes and special character attributes. The distinct character attributes can measure the number of distinct characters in each component of the decomposed email address. The special character attributes can measure the number of special characters (i.e., punctuation or other nonalphanumeric characters) in each component. In some examples, the special character attributes can be further specified to identify specific non-alphanumeric characters to provide separate weighting of such characters. For instance, special characters including“*”, and “!” may each be separately tracked as a specific special character type in generating attributes. Practically, some non-alphanumeric characters, such as periods, underscores, and hyphens may traditionally be more present, while non-traditional special characters can include pound symbols, percentage symbols, carets, ampersands, exclamation marks, and plus symbols. Such non-traditional special characters may thus be separately tracked to provide for separate weighting of such characters as attributes
[0051] A seventh email attribute type can include inner digit number attributes. Inner digit attributes count the consecutive number of digits and numerals when adjacent to alphabetical or punctuation characters in each of the components. In the example of15US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO “John.Doel23@test-fake.mail.com”, the full email address contains a single inner digit number of value 3 to account for “123.” However, the handler component would not have an inner digit number as the handler component terminates with “123”.
[0052] Eighth and ninth email attribute types can include edit distance (i.e. , Levenshtein edit distance) attributes, including handler-name distances and domain distances. Handlername distances can compare identified name components within the handler (e.g., full names, nicknames, initials) within the handler, and compare the handler name components against a database including names communicatively coupled to the disposable email identification computing system 101. For example, the disposable email identification computing system 101 may be communicatively coupled to one or more other repositories storing entity PII including entity first names, middle names, and last names. Via accessing the PII repositories, the email evaluation application 110 can compare handler name components to names stored in the PII repository to determine edit distance scores.
[0053] Because the set of possible domain names is relatively more finite than potential handler names, email attribute ty pes can include domain distances. Domain distances can track the distance of the domain against one or more of a set of common domains. The set of common domains can include those identified as most frequent within the email repository, or can be manually configured. For instance, manual configuration can include incorporating email addresses such as “outlook.com”, “gmail.com”, “yahoo.com”, “hotmail.com”, “aol.com”, “icloud.com”, “gmx.com”, “outlook.com”, “zoho.com”, “msn.com”,“ mail.com”, live.com” and the like.
[0054] Additional email attribute types may be configured through machine-learning techniques. For instance, rule-based attributes can be machine-learning generated according to various machine-learning techniques. In one example the above described attributes and components can be applied to a Skope rules model, configured to output a set of rule-based attributes. The Skope rules model can similarly be provided a supervised, labeled training set of disposable emails 120 and non-disposable emails 122 as accessed from the email address repository 118. The rule-based attributes may then include one or more rule sets determined by the Skope rules model to be most effective in determining whether an email address is disposable or non-disposable. Other rule generation modeling techniques including isolation forest modeling and robust random cut forest modeling may be used in addition or in alternative to Skope rules modeling to produce the rule-based attributes. The rule-based 16US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO attributes, in some instances, can overlap with one or more of the above described attributes. For example, a rule-based attribute could be output as a binary7value indicating whether the depth of a domain is greater than two but less than five, and the distinct character attribute in the handler is greater than three.
[0055] At block 208, the process 200 involves determining a risk score indicating a likelihood that the input email address includes a disposable email address based on applying the set of email attributes to a machine-learning model 114 trained on an email based training data set, where the email-based training data set includes non-disposable email addresses and disposable email addresses. Disposable email addresses can include email addresses lacking PII associated with the user who created or uses the email address. Similarly misleading PII, such as the use of fake names (e.g., “John Doe” email addresses), where the PII does not match the associated user, are indicative of disposable email addresses. Disposable email addresses can also include email addresses which are of single or limited use, where activity patterns associated with the disposable email address are limited to account creation or account access among various distributed computing environments. Non-disposable email addresses in contrast refer to those email addresses reflecting the associated’ user’s intent to maintain consistent use of the email address. Factors indicating anon-disposable email address can include the amount of PII in the email address, and a more routine but diverse activity7pattern associated with the email address (e.g., frequent log-ins, and used for purposes beyond account creation and maintenance). The machine-learning model 114 can include a classification model trained to generate risk scores, where the risk scores indicate a likelihood the input email address is a disposable or non-disposable email address based on the provided email attributes. In some examples, the machine-learning model 114 can include an XGBoost classification model. Operations for training the machine-learning model 114 to provide accurate assessments of email disposability and risk scores are discussed further with respect to FIG. 4.
[0056] At block 210. the process 200 involves transmitting, by the risk assessment computing system and to a remote computing device, a responsive message including at least the input email address and access information based on the risk score for use in controlling access of the entity to an interactive computing environment. The access information can include the risk score. In such examples, a user looking to evaluate a given email address can provide the input email address to the disposable email identification computing system 10117US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO and receive a response providing the risk score, allowing the user to further evaluate the email and determine whether to allow, deny, or restrict access, to a secure computing resource, to an entity associated with the input email address. In other examples, the access information can include an approval or denial token for automatically causing the associated entity ’s access to the secure computing resource to be denied.
[0057] In the examples where the access information includes approval and denial tokens, the risk score can be compared against a risk threshold. The risk threshold can be a maximum risk score allowed for the associated entity’ to access a secure resource. In some examples, the risk threshold can be set by an external entity, such as a privileged user with access to the disposable email identification computing system 101, in the same or other examples, client devices can access the disposable email identification computing system 101 can provide a set of risk score thresholds for client specific secure resources. In response to the risk score being under the risk thresholds, tokens can be set. by the email evaluation application 110. to include an approval token with the email address. Meanwhile, exceeding the risk threshold can cause the email evaluation application 110 to associate a denial token with the input email address. The approval and denial token can be included within the access information as the token is transmitted to the remote computing device. In response, the remote computing device can automatically control access to a secure computing resource based on the provided token.Example Operations for Training a Machine-Model to Determine Email Disposability
[0058] The described operations for assessing the risk whether an email is disposable can include applying generated email attributes to a machine-learning model 114 as described with respect to FIG. 4. Techniques described herein provide means for training the machinelearning models to provide accurate output of the risk scores. FIG. 4 provides an example set of operations for training a machine-learning model to determine email disposability’ risk scores, according to certain examples. For illustrative purposes, the process 400 is described with reference to implementations described above with respect to one or more examples described herein. Other implementations, however, are possible. In some aspects, the operations in FIG. 4 may be implemented in program code that is executed by one or more computing devices such as the model training server 108 depicted in FIG. 1. In some aspects of the present disclosure, one or more operations shown in FIG. 4 may be omitted or performed in a different order. For instance, the first stage operations (402-404) may be 18US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO performed without performing second stage operations (404-406). Similarly, alternative steps to the first stage operations (not shown) may be applied such that only the second stage operations are performed as described with respect to FIG. 4.
[0059] Blocks 402-406 represent a first training stage for training the machine-learning model 114 to output risk scores, while blocks 406-408 discuss a second stage of re-training the machine-learning model 114 to fine-tune the machine-learning model 114 and provide improved accuracy in assessing email disposability7risk scores.
[0060] Turning to the first stage, at block 402, the process 400 involves sampling an email repository to generate an initial data set including non-disposable email addresses and disposable email addresses. Sampling may include stratified sampling of separate repositories including previously labeled disposable emails 120 and non-disposable emails 122 respectively. Alternatively, a full email address repository 118 may be randomly sampled without specific sampling of the segregated disposable emails and non-disposable emails. The retrieved, sampled emails may then be categorized into non-disposable and disposable email addresses. Various other sampling techniques may be employed.
[0061] The initial data set, particularly when generated according to randomly sampling the email address repository 118 as a whole, may result in a heavily unbalanced training set, grossly favoring the non-disposable emails. As a result, training machine-learning models on the initial data set may produce models ill-equipped at detecting and assessing the disposability of a given email address. Machine-learning model accuracy can thus be improved by further manipulating the training set according to further down-sampling techniques. At block 404, the process 400 involves down-sampling the initial data set into a reduced data set such that a ratio of the non-disposable email addresses to the disposable email addresses is under a threshold ratio value. The threshold ratio value can be set to 100: 1 non-disposable to disposable email addresses, in another example to 50:1, and in a further example to 20: 1 threshold ratio values. By down-sampling the initial training data set until the reduced set is under such a threshold value, the machine-learning model 114 may be trained to detect more nuances from a relatively larger set of disposable email addresses.
[0062] At block 406, the process 400 involves training a machine-learning model using the reduced data set. The machine-learning model 114 can generally include decision trees, random forests, logic regression models or other classifier models. Various machine-learning 19US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO model training techniques can be applied such as using stochastic learning with backpropagation, relying on the gradient of a mathematical loss function to adjust the weights of the network. Generally, relying on labeled data sets of disposable emails and nondisposable emails can include supervised learning techniques. According to other examples, unsupervised learning techniques via clustering unlabeled email addresses to identify patterns can be employed as an additional or alternative means for email disposability determinations based on the described attributes. Similarly, semi-supervised learning techniques can be applied where only a portion of the training data set is retrieved from labeled repositories. Training the machine-learning models may also include training according to a variety of libraries such as extreme Gradient Boosting (“XGBoosf ’), Light Gradient Boosting Machine (“LightGBM”) and the like. The machine-learning model hyperparameters may also be tuned according to a variety of techniques including through application of optimization algorithms such as Tree of Parzen Estimators (“TPE”).
[0063] Subsequent to the operations of blocks 402-406, the machine-learning model 114 is trained according to a first stage where the machine-learning model 114 can assess the disposability of email addresses. In some approaches, a second training stage can be applied to the machine-learning model 114 to fine-tune prediction operations. Blocks 406-408 illustrate example operations for fine-tuning the machine-learning model 114.
[0064] At block 408, the process 400 involves identifying a subset of the set of email attributes based on relative performance of the subset. Relative performance and contribution of the attributes can be monitored according to various techniques. In one example, feature sensitivity algorithms, such as the Shapley Additive Explanations (’ SHAP") framework can be applied where SHAP scores are generated for each attribute and indicate the contributions of each attribute with respect to machine-learning model output. Each attribute may then be ranked according to a respective SHAP score, and the subset of the attributes identified based on the ranked SHAP scores. For instance, a threshold number of attributes may be separated from other attributes in the set identified as lower performing. Additionally or alternatively, minimum SHAP scores may be applied to filter out lower-scoring attributes from the set of email attributes in forming the subset.
[0065] In some examples, in addition to identify ing new attributes to weigh and retrain the machine-learning model 114 in a second stage, preparing the second stage training of the machine-learning model 114 can include creating a second training set of disposable and non- 20US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO disposable emails. For example, a second training set of disposable and non-disposable emails can include the first training set, in addition to email addresses that were incorrectly detected as non-disposable email addresses by the machine-learning model 114 trained during the first training stage.
[0066] At block 410, the process involves retraining the machine-learning model 114 using the subset of email attributes. Retraining the machine-learning model 114 during the second stage can include the same techniques described with respect to the first training stage described per block 406, but with the training data adjusted per block 408. By retraining the machine-learning model 114 on the subset of email attributes, the predictive power of the machine-learning model 114 can be improved. For instance, attributes which may tend to be present in larger sets of disposable email addresses (e.g. having high scoring inner-digit attributes) may be identified per block 406 to have greater predictive weight in identifying disposable email addresses, or may be determined to have decreasing predictive weight over the course of a monitoring period. In response, by retraining the machine-learning model 114 during the second stage operations according to operations 406-410, the machine-learning model 114 can react to the predictive weight or changes in predictive weight of each attribute to provide more accurate predictions.
[0067] Similarly, in instances where false positives are detected (i.e.. where the machinelearning model 114 trained according to first stage operations incorrectly identifies an email as disposable), multiple training stages with multiple training sets can improve accuracy. For instance, training the machine-learning model 114 in the second stage with a second training set of disposable and non-disposable emails including the first training set, in addition to email addresses that were incorrectly detected as disposable email addresses can reduce inaccurate predictions by the machine-learning model 114. Such a combination of refining the attributes evaluated by the model, in addition to the ty pes of data used to train the model can provide a robust email address disposability detection model that can detect disposable emails while limiting the number of false positives.Example of Computing, System for Machine-Learning Operations
[0068] Any suitable computing system or group of computing systems can be used to perform the operations for the machine-learning operations described herein. For example, FIG. 5 is a block diagram depicting an example of a computing device, which can be used to 21US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO implement operations performed by the disposable email identification computing system, according to certain examples. The computing device 500 can include various devices for communicating with other devices in the operating environment, as described with respect to FIG. 1. The computing device 500 can include various devices for performing one or more transformation operations described above with respect to FIGS. 1-4.
[0069] The computing device 500 can include a processor 502 that is communicatively coupled to a memory 504. The processor 502 executes computer-executable program code stored in the memory' 504, accesses information stored in the memory 504, or both. Program code may include machine-executable instructions that may represent a procedure, a function, a subprogram, a program, a routine, a subroutine, a module, a software package, a class, or any combination of instructions, data structures, or program statements. A code segment may be coupled to another code segment or a hardware circuit by passing or receiving information, data, arguments, parameters, or memory contents. Information, arguments, parameters, data, etc. may be passed, forw arded, or transmitted via any suitable means including memory sharing, message passing, token passing, network transmission, among others.
[0070] Examples of a processor 502 include a microprocessor, an application-specific integrated circuit, a field-programmable gate array, or any other suitable processing device. The processor 502 can include any number of processing devices, including one. The processor 502 can include or communicate with a memory 504. The memory 504 stores program code that, when executed by the processor 502, causes the processor to perform the operations described in this disclosure.
[0071] The memory 504 can include any suitable non-transitory computer-readable medium. The computer-readable medium can include any electronic, optical, magnetic, or other storage device capable of providing a processor with computer-readable program code or other program code. Non-limiting examples of a computer-readable medium include a magnetic disk, memory chip, optical storage, flash memory, storage class memory. ROM, RAM, an ASIC, magnetic storage, or any other medium from which a computer processor can read and execute program code. The program code may include processor-specific program code generated by a compiler or an interpreter from code w ritten in any suitable computer-programming language. Examples of suitable programming language include Hadoop, C, C++, C#, Visual Basic, Java, Python, Perl, JavaScript, ActionScript, etc.22US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO
[0072] The computing device 500 may also include a number of external or internal devices such as input or output devices. For example, the computing device 500 is shown with an input / output interface 508 that can receive input from input devices or provide output to output devices. A bus 506 can also be included in the computing device 500. The bus 506 can communicatively couple one or more components of the computing device 500.
[0073] The computing device 500 can execute program code 514 that includes the email evaluation application 110 and / or the model training application 112. The program code 514 for the email evaluation application 110 and / or the model training application 112 may be resident in any suitable computer-readable medium and may be executed on any suitable processing device. For example, as depicted in FIG. 5, the program code 514 for the email evaluation application 110 and / or the model training application 112 can reside in the memory 504 at the computing device 500 along with the program data 516 associated with the program code 514, such the email evaluation application 110 and the machine-learning model(s) 114. Executing the email evaluation application 110 or the model training application 112 can configure the processor 502 to perform the operations described herein.
[0074] In some aspects, the computing device 500 can include one or more output devices. One example of an output device is the network interface device 510 depicted in FIG. 5. A network interface device 510 can include any device or group of devices suitable for establishing a wired or wireless data connection to one or more data networks described herein. Non-limiting examples of the network interface device 510 include an Ethernet network adapter, a modem, etc.
[0075] Another example of an output device is the presentation device 512 depicted in FIG. 5. A presentation device 512 can include any device or group of devices suitable for providing visual, auditory, or other suitable sensory output. Non-limiting examples of the presentation device 512 include a touchscreen, a monitor, a speaker, a separate mobile computing device, etc. In some aspects, the presentation device 512 can include a remote client-computing device that communicates with the computing device 500 using one or more data networks described herein. In other aspects, the presentation device 512 can be omitted.General Considerations
[0076] Numerous specific details are set forth herein to provide a thorough understanding of the claimed subject matter. However, those skilled in the art will understand that the 23US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO claimed subject matter may be practiced without these specific details. In other instances, methods, apparatuses, or systems that would be known by one of ordinary7skill have not been described in detail so as not to obscure claimed subject matter.
[0077] Unless specifically stated otherwise, it is appreciated that throughout this specification that terms such as ‘'processing,” '‘computing,” “determining,” and “identifying” or the like refer to actions or processes of a computing device, such as one or more computers or a similar electronic computing device or devices, that manipulate or transform data represented as physical electronic or magnetic quantities within memories, registers, or other information storage devices, transmission devices, or display devices of the computing platform.
[0078] The system or systems discussed herein are not limited to any particular hardware architecture or configuration. A computing device can include any suitable arrangement of components that provides a result conditioned on one or more inputs. Suitable computing devices include multipurpose microprocessor-based computing systems accessing stored software that programs or configures the computing system from a general purpose computing apparatus to a specialized computing apparatus implementing one or more aspects of the present subject matter. Any suitable programming, scripting, or other type of language or combinations of languages may be used to implement the teachings contained herein in softw are to be used in programming or configuring a computing device.
[0079] Aspects of the methods disclosed herein may be performed in the operation of such computing devices. The order of the blocks presented in the examples above can be varied, for example, blocks can be re-ordered, combined, or broken into sub-blocks. Certain blocks or processes can be performed in parallel.
[0080] The use of “adapted to” or “configured to” herein is meant as open and inclusive language that does not foreclose devices adapted to or configured to perform additional tasks or steps. Additionally, the use of “based on” is meant to be open and inclusive, in that a process, step, calculation, or other action “based on"’ one or more recited conditions or values may, in practice, be based on additional conditions or values beyond those recited. Headings, lists, and numbering included herein are for ease of explanation only and are not meant to be limiting.24US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO
[0081] While the present subject matter has been described in detail with respect to specific aspects thereof, it will be appreciated that those skilled in the art, upon attaining an understanding of the foregoing, may readily produce alterations to, variations of, and equivalents to such aspects. Any aspects or examples may be combined with any other aspects or examples. Accordingly, it should be understood that the present disclosure has been presented for purposes of example rather than limitation, and does not preclude inclusion of such modifications, variations, or additions to the present subject matter as would be readily apparent to one of ordinary’ skill in the art.25US200831330011 1
Claims
PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO ClaimsWHAT IS CLAIMED IS:
1. A non-transitory computer-readable storage medium having program code executable by a processing device to perform operations comprising:receiving, by a risk assessment computing system, an input email address associated with an entity;decomposing the input email address into a decomposed email data structure comprising a plurality of components;generating a set of email attributes based on the plurality of components;determining a risk score indicating a likelihood that the input email address comprises a disposable email address based on applying the set of email attributes to a machine-learning model trained on an email-based training data set, wherein the email-based training data set includes nondisposable email addresses and disposable email addresses; andtransmitting, by the risk assessment computing system and to a remote computing device, a responsive message including at least the input email address, and access information based on the risk score for use in controlling access of the entity to an interactive computing environment.
2. The non-transitory computer-readable storage medium of claim 1, wherein receiving tire input email address further includes receiving a request to access a resource within the interactive computing environment, and wherein the operations further comprise:in response to determining the risk score exceeds a threshold value, causing the risk assessment computing system to transmit a denial token denying the request to access the interactive computing environment.
3. The non-transitory computer-readable storage medium of claim 1, wherein the plurality of components includes a handler, a server name, and a domain including a prefix and a suffix.
4. The non-transitory computer-readable storage medium of claim 3, wherein the set of email attributes comprises length attributes identifying a number of characters in the domain or the handler, depth attributes measuring a number of characters separated by punctuation within the domain or handler, ratio attributes of the domain or the handler, or any combination thereof.
5. The non-transitory computer-readable storage medium of claim 3, wherein the set of email attributes comprises handler-name distances, domain distances, or a combination thereof, wherein the handler-name distances are determined based on a comparison of an edit distance between the handler 26US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO and a set of names stored in a data repository communicatively coupled to the risk assessment computing system.
6. The non-transitory computer-readable storage medium of claim 1, wherein the email-based training data set used to train the machine-learning model is generated through steps comprising: sampling an email repository’ to generate an initial data set comprising non-disposable email addresses and disposable email addresses; anddown-sampling the initial data set into a reduced data set such that a ratio of the nondisposable email addresses to disposable email addresses in the reduced data set is under a threshold ratio value.
7. The non-transitory computer-readable storage medium of claim 1, wherein the machinelearning model is trained in multiple stages including a first stage and a second stage, wherein the first stage includes training the machine-learning model on the email-based training data set, and wherein training the machine-learning model in the second stage comprises:identifying a subset of the set of email attributes based on a relative performance of the subset; andretraining the machine-learning model using the subset.
8. A method that includes one or more processing devices performing operations comprising:receiving, by a risk assessment computing system, an input email address associated with an entity;decomposing the input email address into a decomposed email data structure comprising a plurality’ of components;generating a set of email attributes based on the plurality of components;determining a risk score indicating a likelihood the input email address comprises a disposable email address based on applying the set of email attributes to a machine-learning model trained on an email-based training data set, wherein the email-based training data set includes nondisposable email addresses and disposable email addresses; andtransmitting, by the risk assessment computing system, and to a remote computing device, a responsive message including at least the input email address and access information based on the risk score for use in controlling access of the entity to an interactive computing environment.
9. The method of claim 8, wherein receiving the input email address further includes receiving a request to access a resource within the interactive computing environment, and wherein the operations further comprise:27US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO in response to determining the risk score exceeds a threshold value, causing the risk assessment computing system to transmit a denial token denying the request to access the interactive computing environment.
10. The method of claim 8, wherein the plurality of components includes a handler, a server name, and a domain including a prefix and a suffix.
11. The method of claim 10, wherein the set of email attributes comprises length attributes identify ing a number of characters in the domain or the handler, depth attributes measuring a number of characters separated by punctuation within the domain or handler, ratio attributes of the domain or the handler, or any combination thereof.
12. The method of claim 10, wherein the set of email attributes comprises handler-name distances, domain distances, or a combination thereof, wherein the handler-name distances are determined based on a comparison of an edit distance between the handler and a set of names stored in a data repository communicatively coupled to the risk assessment computing system.
13. The method of claim 8, wherein the email-based training data set used to train the machinelearning model is generated through steps comprising:sampling an email repository to generate an initial data set:categorizing the initial data set into non-disposable email addresses and disposable email addresses; anddown-sampling the initial data set into a reduced data set such that a ratio of the nondisposable email addresses to disposable email addresses in the reduced data set is under a threshold ratio value.
14. The method of claim 8, wherein the machine-learning model is trained in multiple stages including a first stage and a second stage, wherein the first stage includes training the machinelearning model on the email-based training data set, and wherein training the machine-learning model in the second stage comprises:identifying a subset of the set of email attributes based on a relative performance of the subset; andretraining the machine-learning model using the subset.
15. A sy stem comprising :a processor; and28US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO a non -transitory computer-readable medium comprising instructions that are executable by the processor to cause the processor to perform operations comprising:receiving, by a risk assessment computing system, an input email address associated with an entity;decomposing the input email address into a decomposed email data structure comprising a plurality of components;generating a set of email attributes based on the plurality of components; determining a risk score indicating a likelihood the input email address comprises a disposable email address based on applying the set of email attributes to a machine-learning model trained on an email-based training data set, wherein the email-based training data set includes non-disposable email addresses and disposable email addresses; and transmitting, by the risk assessment computing system and to a remote computing device, a responsive message including at least the input email address and access information based on the risk score for use in controlling access of the entity to an interactive computing environment.
16. The system of claim 15, wherein receiving the input email address further includes receiving a request to access a resource within the interactive computing environment, and wherein the operations further comprise:in response to determining the risk score exceeds a threshold value, causing the risk assessment computing system to transmit a denial token denying the request to access the interactive computing environment.
17. The system of claim 15, wherein the plurality of components includes a handler, a server name, and a domain including a prefix and a suffix.
18. The system of claim 17, wherein the set of email attributes comprises length attributes identifying a number of characters in the domain or the handler, depth attributes measuring a number of characters separated by punctuation within the domain or handler, ratio attributes of the domain or the handler, or any combination thereof.
19. The system of claim 17, wherein the set of email attributes comprises handler-name distances, domain distances, or a combination thereof, wherein the handler-name distances are determined based on a comparison of an edit distance betw een the handler and a set of names stored in a data repository' communicatively coupled to the risk assessment computing system.29US200831330011 1PATENT Attorney Docket No 096923-1494014 Client Ref. No. EFX-214WO 20. The system of claim 15, wherein the email-based training data set used to tram the machinelearning model is generated through steps comprising:sampling an email repository to generate an initial data set;categorizing the initial data set into non-disposable email addresses and disposable email addresses; anddown-sampling the initial data set into a reduced data set such that a ratio of the nondisposable email addresses to disposable email addresses in the reduced data set is under a threshold ratio value.30US200831330011 1
