System and method for entity network analysis
By analyzing the association level and association growth rate of entity networks, synthetic identities are identified, solving the problem of difficulty in detecting synthetic identities in existing technologies, and achieving efficient and accurate fraud identification and prevention.
Patent Information
- Application Number
- CN201910627379.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2018-07-12
- Filing Date
- 2019-07-12
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2039-07-12
AI Technical Summary
Existing technologies struggle to effectively identify synthetic identities, which may appear legitimate but are difficult to verify, making it challenging to detect and prevent fraudulent activities.
Synthetic identities are identified by analyzing the association levels and growth rates of entity networks. This method involves selecting entities from a group of entity representations, identifying network chains, determining synthetic identities based on the number of associations at the association level, and performing data processing and analysis using computer systems and specialized software.
It improves the accuracy and efficiency of identifying synthetic identities, reduces false alarm rates, and can effectively detect and prevent fraudulent activities.
Smart Images

Figure CN110781229B_ABST
Abstract
Description
Technical Field
[0001] Embodiments described herein relate to systems and methods for identifying fraudulent or synthetic identities based on the growth rate of a network of related entities. Background Art
[0002] Many businesses and / or government agencies face increasing problems related to false identities. Identity fraud based on stolen or manipulated identity information can be difficult to detect. For example, verifying a person's identity may involve obtaining information from independent sources (e.g., credit bureaus, public records, etc.) and verifying that information using data provided by the person. For example, when a person applies to open an account with a business, the business may utilize an identity verification service to check the person's documents against public and private databases to verify the person's identity. In some cases, the applicant's name and other identifying information may be checked against available fraud databases to understand any past fraudulent activity by the applicant.
[0003] Fraudsters are now generating "synthetic identities" that may appear legitimate but are difficult to verify. Because the identity is not real, there may be no reported history or fraudulent activity associated with the synthetic identity. Fraud schemes involving synthetic identities are likely to continue to increase, especially without detection and prevention mechanisms. Therefore, effective methods are needed to detect whether an identity is synthetic. Summary of the invention
[0004] Some or all of the above needs may be addressed by certain embodiments of the present invention.
[0005] According to an exemplary embodiment of the present invention, a computer-implemented method is provided, the method being executed by at least one processor configured to access an electronic database, for identifying one or more synthetic identities in a group of entity representations in the electronic database. Each entity representation in the electronic database may include a plurality of records, each record may include a plurality of fields, at least a first field of the plurality of fields being capable of indicating an association with a similar entity representation in the group. The method may include selecting a first entity representation from the group of entity representations, and for the selected first entity representation, the method includes identifying one or more network chains formed by one or more connections between the first entity representation, a second entity representation, and zero or more intermediate entity representations; and determining an association number for each association level of a predetermined level in at least one of the identified one or more network chains. The method includes: determining an indication that one or more of the first entity representation, the second entity representation, and the zero or more intermediate entity representations are synthetic identities based on the association number of each association level; and outputting entity representation information of one or more entities in the group indicated as synthetic identities.
[0006] In an exemplary embodiment, a system is provided. The system includes at least one memory for storing data and computer executable instructions and at least one processor, the processor being configured to access the at least one memory and further configured to execute the computer executable instructions to identify one or more synthetic identities in a group of entity representations in an electronic database. Each entity representation in the electronic database may include multiple records, each record may include multiple fields, and at least a first field of the multiple fields can indicate an association with a similar entity representation in the group. The one or more synthetic identities are identified by a process of selecting a first entity representation from the group of entity representations, and for the selected first entity representation, the process includes identifying one or more network chains formed by one or more associations between the first entity representation, a second entity representation, and zero or more intermediate entity representations; and in at least one of the identified one or more network chains, determining an association number for each association level of a predetermined level. The process includes: determining an indication that one or more of the first entity representation, the second entity representation, and the zero or more intermediate entity representations are synthetic identities based on the association number of each association level; and outputting entity representation information of one or more entities in the group indicated as synthetic identities.
[0007] These and other embodiments of the invention will be described in more detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0008] Attached Figure 1 According to an exemplary embodiment of the present invention, an example entity network populated with real entities and synthetic identities is shown with associated illustrative network geometry growth according to association levels.
[0009] Attached Figure 2A The number of associations for each association level is shown for natural persons, possible immigrants, and synthetic identities.
[0010] Attached Figure 2B The correlation growth trend of each correlation level for natural persons, possible immigrants and synthetic identities is shown, which corresponds to the following Figure 2A The number of associations at each level is shown.
[0011] Attached Figure 2C The association growth for each association level, eg, natural persons, possible immigrants, and composite identities, is shown.
[0012] Attached Figure 3 According to an exemplary embodiment of the present invention, a representative computer architecture is shown.
[0013] Attached Figure 4According to an exemplary embodiment of the present invention, a block diagram of a process for calculating a growth rate of an entity network is shown.
[0014] Attached Figure 5 According to an exemplary embodiment of the present invention, a flow chart of a method is shown.
[0015] Attached Figure 6 According to an exemplary embodiment of the present invention, a flow chart of another method is shown. DETAILED DESCRIPTION
[0016] In order to facilitate understanding of the principles and features of the present invention, various illustrative embodiments are explained below. The system and method technology described below is intended to be illustrative rather than restrictive. Many suitable end-use applications that can utilize the same or similar functions as described herein are intended to be included within the scope of the present invention.
[0017] In short, various embodiments of the present invention are used to identify synthetic identities in a network of entities by determining the relative geometric growth and / or growth rate of the network of entities according to the level of association.
[0018] As used herein, the term "entity network" is intended to include system diagram information, which may represent direct and / or indirect relationships between entities shown in the diagram. Such information used to determine system diagram information may be obtained or derived from public, private, and / or proprietary sources. Certain exemplary embodiments of the present invention may emphasize public record information, such as shared assets, cohabitation, and other similar types of records that are difficult to fabricate.
[0019] Information from certain social networking platforms (such as Facebook, Instagram, Twitter, etc.) may be considered in some implementations, but the information obtained from these platforms may be weighted according to the reliability of the information. In some exemplary embodiments, information from these social networking platforms may not be used at all.
[0020] As used herein, the terms "degree," "degree of separation," "level of association," or "n-level" as applied to entity network associations may generally refer to a visualized "distance" between two entity representations. In terms of an entity graph of nodes, where each node represents an entity connected to another entity by an edge, a "first-level" relationship is defined herein as two nodes directly connected by an edge without intermediate nodes. This first-level association may be defined by factors such as cohabitation, shared assets, members of the same household, and / or similar factors. As used herein, a "second-level" relationship may refer to an indirect relationship between two nodes or entities of an entity network, where one or more intermediate first-level relationships provide an association linking the two nodes or entities. Typically, an "n-level" relationship may refer to a relationship between two nodes of an entity network graph, where n-1 intermediate first-level relationships provide the association linking the two nodes through n intermediate edges.
[0021] According to certain exemplary embodiments of the present invention, a relationship between two entities may be established through active and / or passive interactions. Active interactions typically involve real people who may be in situations that are difficult to fabricate. For example, associations based on active interactions may include shared assets, shared insurance policies, real people who are jointly named in bankruptcy, etc. Examples of associations based on passive interactions include: establishing a link between two entities through a record of cohabitation. In certain exemplary embodiments, associations based on such passive interactions may be established and / or considered at a predetermined time period, such as after 6 months of cohabitation.
[0022] Because synthetic entities are non-autonomous, they rely on actions performed by a real person to create the "appearance" of certain network associations with other entities. For example, a fraudster who cultivates a non-autonomous identity may have many associations at the first level, but have lower than average associations at higher levels due to the passive nature of the associations. In some cases, the fraudster may easily create 1-level associations between synthetic entities, and only higher-degree associations involving synthetic entities are created through the fraudster's actions. Therefore, according to certain exemplary embodiments of the present invention, synthetic entities may be detected by determining entities that have higher than normal associations at the 1st level, but lower than normal associations at higher levels.
[0023] Attached Figure 1 An exemplary entity network 100 is shown populated with synthetic identities S1-S3 (shown by light nodes) and real entities R1-R7 (shown by dark nodes) with relevant associations 103 between them. Figure 1 Also included is illustration 106 showing the number of connections as a function of separation for certain network connections involving real entities. Illustration 112 shows the number of connections as a function of separation for certain network connections involving synthetic identities according to an exemplary embodiment of the present invention.
[0024] As attached Figure 1 As shown in the exemplary entity network 100 of FIG. 1 , a real entity 102 (shown by R1) can sequentially generate synthetic identities S1-S3 over time. For example, a first synthetic identity S1 can be generated with a pseudonym, but with other identification information (such as an address) that can identify the association between the synthetic identity S1 and the real entity R1. Later, the real entity R1 can generate another synthetic identity 108 (shown by S2) that also has identification information that identifies the association between the synthetic identity S2 and the real entity R1.
[0025] Attached Figure 1 Also shown are legal branches of the entity network 100 involving real entities. For example, one real entity R1 is shown associated to another real entity 104 (shown by R2). For example, R2 may be a spouse or relative of R1, and the two real entities 102, 104 may be associated 103 by a cohabiting address. Subsequently, other associations may be formed by linking R2 and R3, R2 and R4 (and so on).
[0026] As shown in illustration 106, branches of an entity network populated by real entities may be characterized by multiple associations at each level that are distinguishable from networks associated with synthetic identities. For example, since synthetic identities do not represent real people, once these synthetic identities are generated, their associated network associations generally do not extend as far as networks associated with real entities. However, in some cases, a real entity (e.g., a fraudster) may generate additional synthetic identities that have what may appear to be 2 levels of association or higher. Figure 1 A scenario is depicted where one real entity R1 generates (as indicated by the dashed line) another synthetic identity 110 (indicated by S3) with identification information linking S3 to S2. However, in practice, it may be difficult to generate a synthetic network populated by synthetic identities and having associations higher than level 1 due to the lack of normal, autonomous association activity characteristic of normal humans. Furthermore, such a synthetic network may be characterized by linear geometric growth, as shown in inset 112. Certain synthetic networks may be characterized by little or no growth, depending on the construction and maintenance of such networks.
[0027] When a fraudster creates a synthetic identity, the synthetic identity is typically only displayed in the network and relies on the actions of the fraudster to be associated with others. As a result, some networks populated with the synthetic identity's second and higher level associations may include people who are also associated with the fraudster. Certain exemplary embodiments of the present invention may use such information to determine a fraudster's identity.
[0028] Certain exemplary embodiments of the present invention may analyze entity network growth from different perspectives to verify the accuracy of information. In certain exemplary embodiments, an entity may be evaluated by: (a) the age of the entity when it first appears in the record; (b) the length of time the entity has been contributing to its entity network fingerprint; and (c) the current age of the entity. For example, a record associated with a particular entity representation may indicate an appearance age of 16, which may be consistent with the typical appearance age of a minor obtaining a driver's license and / or being employed for the first time. Other records associated with the entity representation may indicate that they have contributed to their entity network record (i.e., fingerprint) for 20 years. Another record may indicate that the current age of the particular entity representative is 26 rather than 36, as indicated by the appearance age and the length of the record's contribution. Certain exemplary embodiments of the present invention may utilize this multi-perspective analysis to erase or mark records in which there are conflicts in the confirmation.
[0029] Attached Figure 2A The number 202 of entity network connections is shown as a function of the level of association 204 for a natural person 206, a possible immigrant 208, and a synthetic identity 210. In the case of the natural person 206, the number of associations 202 is generally substantially higher than the number associated with the synthetic identity 210, particularly for those associations above level 2. Certain exemplary embodiments of the present invention can also distinguish between possible immigrants 208 and synthetic identities 210.
[0030] Attached Figure 2B The trend of association growth 212 for each association level 204 is shown for natural persons 206, possible immigrants 208, and synthetic identities 210, which corresponds to the following: Figure 2A The number of associations per level is shown. In this example, the growth of the natural person 206 (ie, the number of associations per level) is significantly higher than the growth of the synthetic identity 210.
[0031] In an exemplary embodiment of the present invention, the geometric growth "rate" r associated with a given degree of associated separation d can be calculated as follows:
[0032]
[0033] Among them, x d is the correlation number at a given separation, x o is the initial association number (ie, level 1 association). In certain exemplary embodiments, the difference (or delta) between the starting and ending growth rates may be calculated as follows:
[0034] SELF.d1:=l.d1;
[0035] SELF.d2:=l.d2;
[0036] SELF.d3:=l.d3;
[0037] SELF.d4:=l.d4;
[0038] t1:=((l.d2 / l.d1))-1;
[0039] t2:=POWER(l.d3 / l.d1,(1 / 2))-1;
[0040] t3:=POWER(l.d4 / l.d1,(1 / 3))-1;
[0041] SELF.r1:=t1;
[0042] SELF.r2:=t2;
[0043] SELF.r3:=t3;
[0044] SELF.r2r1:=t2 / t1;
[0045] SELF.r3r2:=t3 / t1;
[0046] SELF.delta:=t3 / t1-t2 / t1;
[0047] Where "SELF.delta" is the difference between the starting and ending growth rates, and "SELF.dx" is the count of unique associations linking the target entity through the shortest or lowest level path, as explained below. According to an exemplary embodiment of the present invention, calculating the delta for an individual may include calculating the associations of unique individuals from the lowest observed level to the 4th level.
[0048] In one exemplary embodiment, the lowest level path may be calculated as follows:
[0049] (A->B->C: A, C, 2) represents the situation where A is associated with B, and B is associated with C, where the abbreviation "A, C, 2" means: "A is associated with C at level 2".
[0050] (A->D->F->G->C: A, C, 4) represents the case where A is associated with D, D is associated with F, F is associated with C, C is associated with G, and G is associated with C, where the abbreviation "A, C 4" represents: "A is associated with C at level 4". According to some exemplary embodiments of the present invention, the third value in the abbreviation represents the number of degrees of separation (i.e., hops) from A to C. According to an exemplary embodiment of the present invention, a lowest level path is determined to calculate how "close" C is to A, and the association number of the lowest level path can be used in the calculation of the difference, as described above.
[0051] Attached Figure 2CThe function of the association level of the association growth rate 212 for each association level 204 is shown for exemplary natural persons 214, 216, exemplary possible immigrants 218, 220, and exemplary synthetic identities 222, 224. According to an exemplary embodiment of the present invention, a threshold value may be used to distinguish between a synthetic identity and a real entity. For example, a growth rate of less than about 1.5 for the 4th association level may indicate a synthetic identity. Similarly, a growth rate of less than about 2.0 for the 5th association level may indicate a synthetic identity. Certain embodiments of the present invention may utilize thresholds for one or more association levels to identify synthetic identities.
[0052] According to certain exemplary embodiments, an entity graph may be determined for each known entity in a population. For example, embodiments of the present invention may be used to determine an entity graph for each of the approximately 280 million known and unique entity representations in the United States. In an exemplary embodiment, the entity network for each of the 280 million entity representations may be determined up to approximately level 4, resulting in approximately 2 billion records. Certain exemplary embodiments of the present invention may calculate the growth rate for each record at each level to determine entity representations that are actual or likely synthetic identities.
[0053] In certain exemplary embodiments, an entity graph of an entity may include individuals who are related in some way, including but not limited to family, business partners, friends, past roommates, etc. In one exemplary embodiment, the area of location, address, or activity may be compared to the current home and / or work address of a person in the individual entity network. In certain exemplary embodiments, the distance between these locations may be calculated. According to one exemplary embodiment, such data may be utilized to identify certain patterns, confidence values, scores, etc. In one exemplary embodiment, all members of the entity network may be considered. In other exemplary embodiments, only individuals with certain association criteria may be considered to reduce computational complexity.
[0054] According to certain exemplary embodiments of the present invention, a network diagram of an entity may be determined based on public record information, including but not limited to: housing records, vehicle records, marriage records, divorce records, hospital records, birth records, death records, court records, property records, employment records, business records, address records, tax records, company records, etc.
[0055] Certain embodiments of the present invention may provide higher accuracy than conventional systems and techniques. For example, law enforcement attempting to identify the perpetrators of certain fraudulent activities (e.g., identity theft) may utilize conventional centralized analysis techniques to examine each event individually. The conventional techniques typically utilize high thresholds to filter a large number of events to be analyzed. In other words, because the data that must be analyzed using conventional techniques is so large, highly suspicious activities may be required to identify individuals or entities associated with the activity. Without high thresholds, conventional techniques may have too many potential fraudulent events to investigate. As a result, entities using conventional techniques often overlook collusion by groups that are able to remain below these high thresholds in terms of certain suspicious activities. The systems and methods disclosed herein can effectively utilize existing data to help verify individuals involved in generating synthetic identities with very low false positive rates.
[0056] In an exemplary embodiment, domains of entities may be identified for analysis. For example, data related to a large number (potentially hundreds of millions) of individuals may be collected for analysis. The data may include the identity of the individual, organization, company, etc. The data may include information such as physical address and / or relationship information. According to certain exemplary embodiments of the present invention, one or more types of relationships between the entities may be collected.
[0057] In certain embodiments, a filtering mechanism may be operated against the network, and those networks with abnormal behavior may be retained. Such filtering may typically utilize graph or network analysis, and this form of query / filtering may utilize a subgraph matching routine or fuzzy subgraph matching. However, subgraph matching routines or fuzzy subgraph matching techniques may be NP (non-deterministic polynomial)-complete, and therefore impractical for analyzing large amounts of data. The most notable feature of NP (non-deterministic polynomial)-complete problems is that there is no solution to them quickly. That is, as the size of the problem increases, the time required to solve the problem using any currently known algorithm increases very quickly. This means that, using any amount of computing power available today, the time required to solve a moderately sized version of many of these problems can easily reach billions or trillions of years. Even if the data set being analyzed may be very large, embodiments of the present invention may be utilized to determine or provide associations between entities.
[0058] According to an exemplary embodiment of the present invention, entities may be identified and may include people, companies, places, objects, virtual identities, etc. In an exemplary embodiment, relationships may be formed in a variety of ways and with a variety of attributes. For example, co-occurrence of values in a common field in a database may be utilized, such as the same last name. Relationships may also be formed using multiple co-occurrences of an entity with one or more other attributes, such as people living at two or more addresses.
[0059] According to an exemplary embodiment, relationships may also be formed based on high recurrence and / or frequency of a common relationship. For example, a record of person X sending emails to person Y greater than N times may indicate a relationship between person X and person Y. In another exemplary embodiment, if person X sends or receives an email to or from person Y, and within a short period of time, person Z sends or receives an email to or from person Y, then a relationship may be implied between person X and person Z.
[0060] According to an exemplary embodiment of the present invention, the relationship between entities may include a combination of Boolean, weighted, directional, non-directional and / or multiple relationships. According to certain exemplary embodiments of the present invention, the clustering of entities may rely on a relationship step. In one embodiment, entities may be related by at least two different relationship types. In one embodiment, the relationship used for network association analysis may be established by checking the weight or strength of the association between entities in certain directions, and is conditional on other relationships including time relationships. For example, in one embodiment, the directional relationship between entities X, Y, and Z may be checked, and if there is a link between Y and Z that occurs (in time) after the link is established between X and Y, the association between X, Y, and Z may be followed up.
[0061] Once the networks are identified, a number of methods may be used to filter the networks. For example, in one embodiment, the networks may be scored. In another embodiment, a threshold may be used to identify networks of interest. According to an exemplary embodiment of the present invention, a model may be used to calculate multiple statistics on each network. In one embodiment, the model may be as simple as determining counts. In another embodiment, the model may detect relationships within a network, such as entities that are related to the centroid of the network and are also related to each other. This analysis may provide a measure of the cohesion of the relationships that exist within the network. According to an exemplary embodiment of the present invention, once the statistics have been calculated, the scores and weights of each network may be used to determine which networks exceed a certain threshold and can be classified as "of interest". According to an exemplary embodiment of the present invention, a heuristic scoring model, such as linear regression, neural network analysis, etc., may be used to complete the weighting and / or scoring of the determined statistics.
[0062] Exemplary embodiments of the present invention may utilize specialized computing systems and custom query languages in the processes described herein to provide meaningful results, as may be required due to the need to track and analyze large amounts of data.
[0063] Certain exemplary embodiments of the present invention provide substantial improvements in computer processing speed, memory utilization, and / or programming languages. These improvements provide certain technical contributions to detecting relationships between individuals. In certain exemplary embodiments, the improved computer system disclosed herein can analyze an entire group, such as all known persons in the United States, and related activities. Through the improvements in computer processing speed, memory utilization, and / or programming languages disclosed herein, the ability to calculate such a large amount of data on the scale required to provide effective outlier detection and information is achieved. Those of ordinary skill in the art can recognize that traditional methods such as human activities, paper and pencil analysis, and even traditional calculations using general-purpose computers and / or off-the-shelf software are not sufficient to provide the data processing level required for effective relationship linking on the expected scale of the present invention. The special-purpose computers and special-purpose programming languages disclosed herein can improve computer speed, memory utilization, and / or reduce disk read / write operations, thereby improving computing technology and realizing the disclosed invention.
[0064] Certain exemplary embodiments of the present invention may be implemented by a dedicated high performance computing and communications (HPCC) system in combination with a dedicated software linking technology called Scalable Automatic Linking Technology (SALT). Scalable Automatic Linking Technology (SALT) and High Performance Computing and Communications (HPCC) are developed and provided by LexisNexis Risk Solutions, Inc. (the assignee of the technology of the present invention). For example, the High Performance Computing and Communications (HPCC) system provides a data-intensive supercomputing platform designed to solve big data problems. As an alternative to distributed computing, the High Performance Computing and Communications (HPCC) platform provides a compatible single architecture that enables efficient processing. The Scalable Automatic Linking Technology (SALT) module, together with the High Performance Computing and Communications (HPCC) system, provides technical improvements in computer processing that facilitate the present invention and provide useful, tangible results that were previously unattainable. For example, certain exemplary embodiments of the present invention can process large data sets that are computationally intensive and require special software and hardware.
[0065] One of the problems that plagued previous "relationship determination" solutions involving massive data sets is the extremely long runtime and large amounts of memory / disk space required. One of the solutions provided by the technology provided by the present invention involves the implementation and efficiency improvements of computer systems and software that process relationship data and provide the required data in a reasonable amount of time. Certain exemplary embodiments of the present invention can be utilized to improve the efficiency of identity-based fraud detection, such as the generation and use of synthetic identities.
[0066] For example, determining the relationship between records may follow the classic n-squared process of time and disk space. According to an exemplary embodiment of the present invention, the Scalable Automatic Linking Technology (SALT) provides a method that can utilize lightweight self-associations, for example, in generating an embeddable common LISP language (ECL). But disk space utilization may still be high. Certain exemplary embodiments of the present invention may enable a core association to be divided into multiple parts, each of which is persistent. This has the advantage of breaking a potentially very long connection into n parts while allowing other time parts. If the final links are quite sparse, this will reduce disk consumption by a factor of n. In terms of performance, it should be noted that if n can be high enough so that the output of each association does not overflow to disk, the relationship calculation process can have significantly faster performance.
[0067] According to certain exemplary embodiments, the linking of records may be performed by certain other specialized programming and analysis software. For example, record linking fits into a general class of data processing known as data integration, which may be defined as the problem of combining information from multiple heterogeneous data sources. Data integration may include data preparation steps, such as parsing, analyzing, cleaning, normalizing, and resolving and standardizing raw input data prior to record linking to improve the quality of the input data and make the data more compatible and comparable (these data preparation steps are sometimes referred to as ETL (extract, transform, load) or extract, transform, load).
[0068] Some details of using Scalable Automatic Linking Technology (SALT) are included in the Appendix section of this application. According to an exemplary embodiment of the present invention, Scalable Automatic Linking Technology (SALT) can provide data profiling and data hygiene applications to support the data preparation process. In addition, Scalable Automatic Linking Technology (SALT) provides a general data ingestion application that allows input files to be combined or merged with an existing base file. Scalable Automatic Linking Technology (SALT) can be used to generate a parsing and classification engine for unstructured data that can be used in data preparation. The data preparation step is typically followed by the actual record linking or clustering process. Scalable Automatic Linking Technology (SALT) provides applications for several different types of record linking, including internal, external, and remote.
[0069] Data profiling, data hygiene, and data source compatibility checks, while key components of the record linkage process, also have their own value in the data integration process and can be fully leveraged by SALT even if record linkage is not a necessary part of a particular data work unit. SALT uses high-level concepts such as term specificity to determine the relevance / weight of specific fields within the scope of the linkage process, as well as mathematical models based on the input data, rather than the need to manually code user rules, which can be key to the overall efficiency of the approach.
[0070] Scalable Automatic Linking Technology (SALT) prevents fraud by verifying identity, address and other factors, and using information about relationships to see where collusive activity may be occurring.
[0071] According to an exemplary embodiment of the present invention, and as described above, a persistent data structure can be used as part of a core association to split, for example, to improve the performance of a computer processor and / or reduce the disk / memory utilization requirements for determining the relationship between records. According to certain exemplary embodiments of the present invention, the persistent data structure is a data structure that retains a previous version of itself when a data structure is modified. Such data structures can be virtually immutable because their operations do not update the structure in place, but can produce a new updated structure. Certain exemplary embodiments can utilize a fusion or merge operation that can create a new version from two previous versions. In certain exemplary embodiments, the persistent data structure can also be created using in-place updates of data, and these data structures can generally use less time or storage space than their purely functional counterparts. In certain exemplary embodiments, persistence can be achieved by simple replication. Certain exemplary embodiments of the present invention utilize the similarity between the old and new versions to share structures between versions.
[0072] Certain embodiments of the invention may facilitate detecting possible, probable and / or actual synthetic identity related fraud, for example, credit, payment or benefit related fraud. Certain exemplary embodiments facilitate disambiguating input information and determining the likelihood of a synthetic identity.
[0073] According to an exemplary embodiment of the present invention, input information associated with an entity network may be processed, weighted, scored, etc., such as to disambiguate the information. For example, some embodiments may utilize one or more input data fields to validate or correct other input data fields.
[0074] In certain exemplary embodiments, data may be received from a first information source associated with an entity. For example, an entity may submit an application for certain benefits, services, credit, etc., and the application may include certain identifying information received from the entity, such as name, social security number, address, etc. This "application information" may be considered to come from the first information source, either directly from the entity or from a supplier, business, government agency, etc. According to an exemplary embodiment of the present invention, independent data from a second information source may be received to check or verify the data provided by the entity received from the first information source. In certain exemplary embodiments, the independent information from the second source is not provided by the entity. However, in certain exemplary embodiments, all or a portion of the information provided by the entity (such as information received from the first information source) may be used at least in part to generate the independent information.
[0075] In an exemplary embodiment, a request for an activity may be used as input data. For example, the request may be for a tax refund. In an exemplary embodiment, the request may include the name, street address, and social security number (SSN) of the requester, wherein the social security number (SSN) has a typographical error (intentional or unintentional). In this example, one or more public or private databases may be searched to find reference records that match the input information. However, because the input social security number (SSN) is incorrect, reference records that match the name and street address may be returned, but with a different associated social security number (SSN). According to certain exemplary embodiments, the input information may be labeled, weighted, scored, and / or corrected based on one or more factors or metrics, including, but not limited to: fields in the reference record with field values that are exact matches, partial matches, non-matches, etc., corresponding field values.
[0076] Exemplary embodiments of the present invention can reduce false positives and increase the likelihood of identifying and preventing fraud based on one or more synthetic identities. According to an exemplary embodiment of the present invention, a model can be utilized to process identity-related input information relative to reference information (e.g., obtained from one or more public or private databases) to determine whether a presented input identity corresponds to a real identity, a correct identity, and / or a possible synthetic identity.
[0077] Certain exemplary embodiments of the present invention may determine or estimate the probability of a synthetic identity based on a set of parameters. In an exemplary embodiment, the parameters may be used to check the input data, such as name, address, and social security number, to determine whether such data corresponds to a real identity. In an exemplary embodiment, the input data may be compared to the reference data, for example, to determine whether field values match, do not match, weight, etc. In certain exemplary embodiments of the present invention, the input data (or related entity records) may be scored to indicate the probability that it corresponds to a real identity.
[0078] In some cases, a model may be utilized to score the input identity elements, e.g., to find flaws in the input data. For example, if the input data is scored as having a sufficiently high probability of corresponding to a true identity, even though there may be certain flaws in the input or reference data, once those flaws are found, the process may disambiguate the data. For example, in one embodiment, the disambiguation may be used to determine how many other identities are related to the input social security number (SSN). According to an exemplary embodiment, controls for relatives may be used to minimize the number of similar records, e.g., that may be due to Jr. and Sr. designations.
[0079] In an exemplary embodiment, the container data can be used to derive a date of birth, for example, based on a matching reference record. In an exemplary embodiment, the derived date of birth can be compared to the date of issuance of a social security number (SSN). If the date of the social security number (SSN) is before the date of birth, a flag can be added to the record as an indicator of fraud.
[0080] Attached Figure 3 According to an exemplary embodiment of the present invention, a block diagram of an exemplary computer system architecture 300 is shown. Various embodiments and methods herein may be embodied in a non-transitory computer-readable medium executed by a processor. It should be understood that the architecture 300 is provided for exemplary purposes only and does not limit the scope of various implementations of communication systems and methods. In certain exemplary embodiments, the system architecture 300 may be embodied in a dedicated high-performance computing and communications (HPCC) system, such as that developed and provided by LexisNexis Risk Solutions, Inc., the assignee of the technology of the present invention. For example, a high-performance computing and communications (HPCC) system provides a data-intensive supercomputing platform designed to solve big data problems. Various implementations and methods herein may be embodied in a non-transitory computer-readable medium executed by a processor.
[0081] Attached Figure 3The illustrated architecture 300 includes a central processing unit (CPU) 302 that processes computer instructions, and a display interface 304 that serves as a communication interface and provides functionality for providing video, graphics, images, and text on the display. In certain exemplary embodiments of the present invention, the display interface 304 may be directly connected to a local display. In another exemplary embodiment, the display interface 304 may be configured to provide data, images, and other information to an external / remote display or computer that is not necessarily connected to the particular central processing unit (CPU) 302. In certain exemplary embodiments, the display interface 304 may, for example, wirelessly communicate with an external / remote display via a Wi-Fi channel or other available network connection interface 312.
[0082] The architecture 300 may include a keyboard interface 306 that provides a communication interface with a keyboard; and a pointing device interface 308 that provides a communication interface to a pointing device, mouse, and / or touch screen. Exemplary embodiments of the architecture 300 may include an antenna interface 310 that provides a communication interface with an antenna; and a network association interface 312 that provides a communication interface with a network. As described above, the display interface 304 may communicate with the network association interface 312, for example, to provide information for display on a remote display that is not directly connected or attached to the system. In some embodiments, a camera interface 314 may be provided that may serve as a communication interface and / or provide functionality for acquiring digital images from a camera. In some embodiments, a sound interface 316 is provided as a communication interface for converting sound into electrical signals using a microphone and converting electrical signals into sound using a speaker. According to an exemplary embodiment, a random access memory (RAM) 318 is provided, wherein computer instructions and data may be stored in a volatile memory device for processing by a central processing unit (CPU) 302.
[0083] According to an exemplary embodiment, the architecture 300 includes a read-only memory (ROM) 320, where unchanged low-level system code or data for basic system functions, such as basic input and output (I / O), booting or receiving keystrokes from a keyboard, are stored in a non-volatile storage device. According to an exemplary embodiment, the architecture 300 includes a storage medium 322 or other suitable type of memory (e.g., random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic disk, optical disk, floppy disk, hard disk, removable cartridge, flash drive), where the files include a stored operating system 324, application programs 326, and data files 328. The application programs 326 may include instructions for organizing, storing, retrieving, comparing and / or analyzing various associations related to attributes and entities related to embodiments of the present invention. According to an exemplary embodiment of the present invention, the analysis system, the clustering unit and / or the scoring unit may be implemented at least in part by the application 326 interacting with data from the read-only memory (ROM) 320 or other memory storage medium 322, and may be activated by interaction of the central processing unit (CPU) 302 and the bus 334 with the operating system 324.
[0084] According to an exemplary embodiment, the architecture 300 includes a power supply 330 that provides appropriate alternating current (AC) or direct current (DC) to the power components. According to an exemplary embodiment, the architecture 300 may include a telephone subsystem 332 that allows the device 300 to send and receive voice over a telephone network. The component devices and the central processing unit (CPU) 302 communicate with each other via a bus 334.
[0085] According to an exemplary embodiment, the central processing unit (CPU) 302 has a suitable structure as a computer processor. In one configuration, the computer central processing unit (CPU) 302 may include more than one processing unit. The random access memory (RAM) 318 is connected to the computer bus 334 to provide fast random access memory (RAM) storage to the central processing unit (CPU) 302 during the execution of software programs such as operating system applications and device drivers. More specifically, the central processing unit (CPU) 302 loads computer executable processing steps from the storage medium 322 or other media into the fields of the random access memory (RAM) 318 to execute software programs. Data can be stored in the random access memory (RAM) 318, where the data can be accessed by the computer central processing unit (CPU) 302 during execution. In an exemplary configuration, the device 300 includes at least 128MB of RAM and 256MB of flash memory.
[0086] The storage medium 322 itself may include multiple physical drive units, such as a redundant array of independent disks (RAID), a floppy drive, a flash memory, a USB flash drive, an external hard drive, a thumb drive, a pen drive, a key drive, a high-density versatile disk (HD-DVD) optical drive, an internal hard drive, a Blu-ray optical drive or a holographic digital data storage (HDDS) optical drive, an external mini dual in-line memory module (DIMM) synchronous dynamic random access memory (SDRAM), or an external micro dual in-line memory module synchronous dynamic random access memory (DIMM SDRAM). Such computer-readable storage media allow the device 300 to access computer-executable processing steps, applications, etc. stored on removable and non-removable storage media to download data from the device 300 or upload data to the device 300. A computer program product, such as a computer program product utilizing a communication system, may be tangibly embodied in the storage medium 322, wherein the storage medium 322 may include a machine-readable storage medium.
[0087] According to an exemplary embodiment of the present invention, the dedicated hardware and instruction set can process a large number of records to detect indicators of fraudulent activity. In some cases, when the initial data build is compiled, the number of records can approach or exceed 3 billion records. According to an exemplary embodiment of the present invention, these records can be processed by Scalable Automatic Linking Technology (SALT) to generate a relative build that can result in more records and relationships. For example, in some cases, the number of records in the relative build can approach or exceed 30 billion records.
[0088] According to an exemplary embodiment of the present invention, the relative construction records may be post-processed to provide a reduced set of records (e.g., approximately 14 billion records). The reason for the reduction in the number of records may be to eliminate duplication, clean up data, correct data errors, delete records with low accuracy or low association credibility, etc. In certain exemplary embodiments, the reduced set of records may include relationship types, relationship history, links, etc. between individual identities. According to an exemplary embodiment, relationship links can be calculated by graph analysis in combination with knowledge engineering language (KEL) and / or scalable automatic linking technology (SALT), as previously described, which can provide certain speed, efficiency and / or memory utilization that is superior to previous calculation languages.
[0089] Certain exemplary embodiments of the present invention may identify errors in data. For example, data provided by an information provider may contain errors that, if not detected, may generate erroneous results. Certain exemplary embodiments of the present invention may be used to measure the accuracy and / or quality of available data, such as by cross-checking to include, erase, correct, or reject the data before deriving entity network connections using the data when fully analyzed. According to an exemplary embodiment of the present invention, such data quality may be determined and / or improved by one or more of cross-checking, erasing to correct errors, and scoring to use or reject the data.
[0090] Attached Figure 4 According to an exemplary embodiment of the present invention, a block diagram of an entity network growth rate calculation process 400 is shown. The process 400 may obtain and / or receive a plurality of data 402 to be analyzed. In certain exemplary embodiments, a portion of the data 402 may be stored locally. In certain exemplary embodiments, all or part of the data may be obtained via one or more local or networked databases 408. According to an exemplary embodiment, the data 402 may be processed 404, and an output 406 may be generated. In an exemplary embodiment, the data 402 may include information about shared assets, cohabitation, and / or other common data that may indicate an association between two entity representations. In certain exemplary embodiments, the data 402 may include activity information 410 associated with some or all of the entity representations, including but not limited to time frames, where a particular address is associated with a corresponding identity.
[0091] In an exemplary embodiment, the process 400 may receive data 402 in various forms and may process 404 the data 402 to derive relationships and associations 412. In certain exemplary embodiments, these relationships may be used to derive entity networks 414 of some or all entity representations. In an exemplary embodiment, the self-consistency of the data used to derive the relationships, associations 412, and entity networks 414 may be checked or verified 416, for example, by comparing the age of appearance, the age of the entity data fingerprint, and the age of the entity. If such data is inconsistent, the relevant records may be marked. Certain exemplary embodiments of the present invention may identify errors in the data. For example, data provided by information vendors may contain errors that, if not detected, may generate erroneous results. Certain exemplary embodiments of the present invention may be used to measure the accuracy and / or quality of available data, such as by cross-checking, so as to include, erase, correct, or reject the data before using the data to derive entity network connections. According to an exemplary embodiment of the present invention, such data quality may be determined and / or improved by one or more of cross-checking, erasing to correct errors, and scoring to use or reject the data.
[0092] According to certain exemplary embodiments of the invention, as previously discussed, the derived entity network 414 may be processed to determine the growth of the network 418. According to an exemplary embodiment of the invention, the determined growth 418 may be processed 406 to generate and output 406, which may include entity attributes 420, such as a likelihood score that the entity is a composite identity, entity characterization information for the composite identity, and / or an indication of a possible creator of the composite identity.
[0093] According to an exemplary embodiment of the present invention, the process 400 can utilize publicly available data, which may include hundreds of millions of records. The process 400 can also clean and standardize the data to reduce the likelihood that matching entities are considered different entities. Before creating a graph, the process 400 can use the data to build a large-scale network of the group in question and its related connections.
[0094] According to an exemplary embodiment, the process 400 may utilize relatively large-scale supercomputing capabilities and analytics to target organized collusion.Exemplary embodiments of the systems and methods disclosed herein may rely on large-scale, dedicated, parallel processing computing platforms to increase the flexibility and scale of the solution.
[0095] Exemplary embodiments of the systems and methods disclosed herein can measure behaviors, activities, and / or relationships to proactively and effectively expose criminal groups and collusion circles. Unlike many traditional systems, the systems and methods disclosed herein are not necessarily limited to activities or circles operating in a single geographic location, and are not necessarily limited to a short period of time. The systems and methods disclosed herein can be used to determine whether the generation of synthetic identities falls into an organized circle or a certain geographic location.
[0096] In an exemplary embodiment, filters may be utilized to reduce the data set to identify groups that demonstrate the greatest association based on a scoring algorithm. In an exemplary embodiment, the systems and methods disclosed herein may utilize scores that match or exceed a predetermined set of criteria to be flagged for evaluation. In an exemplary embodiment of the invention, filtering may utilize one or more target scores, which may be selected based on the scoring algorithm. In an exemplary embodiment, geographic entity networks having a score greater than or equal to a target score may be flagged as potential collusion.
[0097] Now refer to the attached Figure 5 , describing an exemplary method 500 for identifying one or more synthetic identities. The method 500 begins in box 502 and, according to an exemplary embodiment, includes selecting a first entity representation from a group of entity representations in an electronic database. In box 504, the method 500 includes identifying one or more network chains formed by one or more associations between the first entity representation, a second entity representation, and zero or more intermediate entity representations. In box 506, the method 500 includes determining an association number for each association level of a predetermined number of levels in at least one of the identified one or more network chains. In box 508, the method 500 includes determining an indication that one or more of the first entity representation, the second entity representation, and the zero or more intermediate entity representations are synthetic identities based on the association number for each association level. In box 510, the method 500 includes outputting entity representation information for one or more entities in the group that are marked as synthetic identities.
[0098] In some exemplary embodiments, determining the association number for each association level may include: determining a shortest network link of the identified one or more network links; and determining the association number for each association level in the shortest network link.
[0099] In certain exemplary embodiments, the indication that one or more of the first entity representation, the second entity representation, and the zero or more intermediate entity representations are composite identities is also determined based on an establishment date of the one or more associations.
[0100] In certain exemplary embodiments, the one or more associations may include one or more of a relative association, a business-related association, a social network association, a shared asset association, and a historical address association.
[0101] According to an exemplary embodiment of the present invention, the one or more associations may be determined using a system diagram, wherein the system diagram includes information associated with one or more public records, the public records including one or more of housing records, vehicle records, marriage records, divorce records, hospital records, death records, court records, property records, employment records, business records, address records, tax records, corporate records, prison records, and utility records, wherein the public utility records include one or more of public utility associations, disconnected associations, and related service addresses.
[0102] According to an exemplary embodiment of the present invention, the one or more associations are based at least in part on distances between geographic areas adjacent to a plurality of locations associated with the entity representations of one or more activities in common.
[0103] Certain exemplary embodiments of the present invention may further comprise determining a geometric growth rate of the one or more associations between the first entity representation, the second entity representation and the zero or more intermediate entity representations, and wherein the indicating is further based on the determined geometric growth rate.
[0104] In certain exemplary embodiments, the indication may also be based on a difference between the determined geometric growth rates.
[0105] According to an exemplary embodiment of the present invention, the entity representation group may include 10 million or more entities.
[0106] In certain exemplary embodiments, the predetermined number of levels is 5 or less.
[0107] Now refer to the attached Figure 6, describing another exemplary method 600 for identifying one or more synthetic identities. The method 600 begins in block 602 and, according to an exemplary embodiment, includes determining one or more records indicating an association of a first entity representation with a second entity representation. In block 604, the method 600 includes identifying one or more network chains formed by at least the first entity representation, the second entity representation, and one or more subsequent entity representations indirectly associated to the selected first entity representation via the second entity representation. In block 606, the method 600 includes determining an association date associated with each association in the one or more identified network chains. In block 608, the method 600 includes determining a growth rate of the one or more identified network chains based on the association date. In block 610, the method 600 includes identifying an indication of a likelihood that one or more of the first entity and the second entity are associated with a synthetic identity based on the determined network chain growth rate. In block 612, the method 600 includes outputting entity representation information of one or more entities indicated as synthetic identities in the population.
[0108] Certain exemplary embodiments of the present invention may further include storing, in a third field of the record associated with the selected first entity representation, an indication of the determined highest growth rate associated with the identified network link including the second entity representation.
[0109] According to an exemplary embodiment of the present invention, determining the one or more records indicating an association of the selected first entity representation with the second entity representation includes determining a lowest level association based at least in part on an earliest date of one or more association establishment dates.
[0110] In an exemplary embodiment, the first-level association between the first entity representation and a second entity representation may include one or more of a relative association, a business-related association, an entity network association, and a historical address association.
[0111] According to an exemplary embodiment of the present invention, a system graph is used to determine the first level association of the first entity representation with the second entity representation. The system graph includes information related to one or more public records, wherein the public records include one or more of housing records, vehicle records, marriage records, divorce records, hospital records, death records, court records, property records, employment records, business records, address records, tax records, corporate records, prison records, and utility records, wherein the utility records include one or more of public facility associations, disconnected associations, and related service addresses.
[0112] In certain exemplary embodiments, the association of the first entity representation with the second entity representation is based at least in part on distances between geographic areas proximate to a plurality of locations associated with one or more activities common to the first and second entity representations.
[0113] According to an exemplary embodiment of the present invention, the growth rate is based on a plurality of associations and dates of the associations.
[0114] Certain exemplary embodiments of the invention may include sorting entity representations by growth rate.
[0115] Certain exemplary embodiments of the invention may include outputting an indication identifying entity representations having a growth rate above a selected threshold.
[0116] According to an exemplary embodiment of the present invention, the group of entity representations may include 10 million or more entities.
[0117] Certain embodiments of the present technology can be used to identify potential organizations for identity theft, fraud, or other activities that can be spread out geographically. For example, the input data for the system diagram can be derived from the address history of the history of the group to be examined, and such address history can be used to link individuals based on, for example, family, residential, and business relationships. Embodiments of the present invention can use such information to determine associations, weight metrics, and / or score data.
[0118] According to exemplary embodiments, certain technical effects may be provided, such as creating certain systems and methods capable of identifying a person associated with various other entities. Embodiments of the present invention may be utilized to examine the growth of networks of entities at various degrees of separation to identify possible synthetic identities. According to exemplary embodiments, certain technical effects may be provided, such as creating certain systems and methods capable of identifying a person who has generated a synthetic identity.
[0119] Without limitation, some of the features considered for determining associations relevant to embodiments of the present invention may include: the number of people who live or work in close proximity in a person's social, family, or business network; the geographic distribution of a person's physical network; and the rate of false representation of identities in a person's network. According to an exemplary embodiment, analysis of these features may allow the system to identify social, family, or business networks, etc., that have a high correlation with fraudulent activity. As previously mentioned, such events may be associated with synthetic identities, but the disclosed embodiments may be applied to other applications that utilize physical network growth rates to identify certain individuals.
[0120] Various embodiments of the systems and methods disclosed herein may be embodied in whole or in part in a computer program product stored in a non-transitory computer-readable medium executed by one or more processors. It should therefore be understood that various aspects of the present invention may include hardware or software of a computer system, as described above with respect to the attached Figure 3 It should also be understood that although these units may be discussed here as being different from each other, they can be implemented in various ways. The distinctions between them throughout the disclosure are only for illustrative purposes and are based on operational distinctions.
[0121] The applications of various embodiments of the systems and methods disclosed herein need not be limited to those described above. For example, one example may be used to identify potential activities related to false representation of credit card applications, identity fraud, investment fraud, crime, commercial fraud, mail and / or delivery fraud, and various other types of activities that may involve geographic information, shared assets, cohabitation, and / or other entity network associations.
[0122] Although certain embodiments of the present invention have been described in conjunction with what are currently considered to be the most practical and various implementations, it should be understood that the technology of the present invention is not limited to the disclosed implementations, but on the contrary, it is intended to cover various modifications and equivalent arrangements included within the scope of the appended claims. Although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.
[0123] This written description uses examples to disclose certain embodiments of the invention, including the best mode, to enable those skilled in the art to practice certain embodiments of the invention, including making and using any device or system and performing any incorporated method. The patentable scope of certain embodiments of the invention is defined in the claims and may include other examples that occur to those skilled in the art. These other examples are intended to fall within the scope of the claims if they do not have structural elements that differ from the literal language of the claims, or if they include equivalent structural elements that are insubstantially different from the literal language of the claims.
[0124] appendix
[0125] Scalable Autolinking Technology (SALT) for relational linking
[0126] Relationship:Relationship name:Basic (field list):Duplicate removal (field list)
[0127] [:score(field list)][:track(field list)multiple result sets(n)li:split(n)]
[0128] [:threshold(n)][:blockthreshold(n)]
[0129] (RELATIONSHIP:relationshipname:BASIS(FieldList):DEDUP(FieldList)
[0130] [:SCORE(FieldList)][:TRACK(FieldList)IMULTIPLE(n)li:SPLIT(n)]
[0131] [:THRESHOLD(n)][:BLOCKTHRESHOLD(n)])
[0132] Relationship:Relationship Name:Relationship List
[0133] [:multiple(n)][:threshold(n)][:blockthreshold(n)]
[0134] [:links(none|all|direct|cross)]
[0135] (RELATIONSHIP:relationshipname:RelationshipList
[0136] [:MULTIPLE(n)][:THRESHOLD(n)][:BLOCKTHRESHOLD(n)]
[0137] [:LINK(NONE|ALL|DIRECT|CROSS)])
[0138] Relationship Name User-specified logical name of the entity relationship to be calculated in the relationship module
[0139] Basis (field list) The field list specified in the field list parameter specifies the basis for the relationship. The field list contains a list of field names separated by colon ':' characters that must be equal between pairs of records. The fields specified in the field list: ? : must be equal to or null (left.field = right.field or left.field = null or right.field = null) (left.field = right.field or left.field = null or right.field = null). The following field: -: indicates that the field must not match. Fuzzy matching of fields is not currently supported.
[0140] deduplication(fieldlist) Specifies a list of field names separated by a colon ':' character to be used for deduplication when matching pairs of records between clusters to compute tie counts. Deduplication prevents double counting when a cluster contains multiple records with equal base fields. Typically, the fieldlist is the same as the fixed part of the base fieldlist. Deduplication fields must be part of the base.
[0141] score(field-list) Specifies an optional list of fields separated by colon ':' characters, which will be scored once per relation. Fields following :-: will be disregarded for calculating the threshold. The score fields cannot be part of the basis.
[0142] tracking(fieldlist) Specifies an optional list of record date fields used to track the dates on which relationship information was collected.
[0143] Multiple (N) specifies the minimum number of entity links that must appear before a relationship is declared.
[0144] Splits(N) Specifies the number of different persistences into which the primary relationship association is split (if not specified, the default value is 1). This can improve performance and reduce disk space utilization.
[0145] Threshold (N) In Scalable Automatic Linking Technology (SALT), relationships are scored the same as internal links, and for a relationship to be declared (counted), the same internal link matching threshold must be exceeded. The default threshold is the global default for internal links. The threshold (n) parameter allows overriding the default.
[0146] Chunk Threshold (N) The Chunk Threshold can be used to override the default Chunk Threshold (5 less than the overall default) for computing relationships. Lower number = more matches and slower times.
[0147] The link(NONE|ALL|DIRECT|CROSS) link parameter is used to define how a given relationship is used to create link candidates for internal linking. The default is LINK(ALL). If LINK(NONE) is specified, the relationship will not participate in internal linking. If LINK(DIRECT) is specified, the relationship will cause both sides of the relationship to be considered as two parts of the same entity. If LINK(CROSS) is specified, then if D1r D2 and D1r D3, then CROSS will cause D2 and D3 to be considered as possible entity link candidates. If LINK(ALL) is specified, then both direct and cross options are evaluated. NOTE: The relationship base will be used to support the fields of the base in the matching logic. So if you have a base of (fname:Iname) - then the name concept will be supported regardless of the base score. This gives a way that if a field match between two entities has 2 (or more) relatively weak values, then the field match will be scored very highly. (For a fuller explanation of support - see the properties file)
[0148] The second form of the relation statement allows a relation to be formed as a sum of other relations. The relation list parameter allows specifying a list of relation names separated by colon ':' characters.
[0149] Scalable Automatic Linking Technology (SALT) internal linking provides the ability to cluster records together to form entities. In some cases, the goal is not to determine whether two records or clusters are close enough to be part of the same entity, but to determine whether there is a statistically significant connection between two clusters and record this relationship. The RELATIONSHIP statement provides this functionality.
[0150] When clustering a single entity using SALT internal linking, there may not be enough information in one record to perform the linking, even if the field values are transferred. SALT internal linking will prevent these records from matching. Relationships provide a way to record instances when multiple occurrences of a specific set of fields (base field list) that match between clusters provide additional clues that a match may exist. For example, while the name match is not strong, the first name, middle name, and two different last names match much stronger between clusters. Relationships cannot currently be used automatically as part of SALT linking, but using the RELATIONSHIP statement SALT can generate an additional relationship file that can be used to create an ATTRIBUTEFILE as input to another SALT process. For example, if you have a relationship between entities ID1 and ID2, the relationship file can be deduplicated and predicted so that and are in the file, and then use the ATTRIBUTEFILE statement to declare one ID as the ID field and the other IDs as value fields. See below for a description of the output format of the relationship file.
[0151] The core part of the SALT relationship is the BASIS; the BASIS is a list of fields that must be the same between two clusters in order for a link to exist. The fields following the :?: field in the BASIS field list must be the same or must be empty. Therefore, the relationship for calculating and weighing the number of shared addresses between two clusters can be:
[0152] RELATIONSHIP:COHABIT:BASIS(PRIMNAME:PRIMRANGE:CITYNAME:ST:?:SECRANGE)
[0153] Note that the second scope comes after the :?:, which defines that they must be equal or must be empty.
[0154] You can also use ATTRIBUTEFILE VALUES fields as part of the BASIS for a relationship.
[0155] As it stands, this will count all pairs of records between two matching clusters, but it may double count if a cluster has two or more copies of an address. To avoid this problem, you can also specify a deduplication (DEDUP) criterion as a list of fields to dedupe. It is normal for the deduplication criterion to be the same as the fixed part of the base. However, with addresses, a road often has multiple names and a city has multiple representations. Therefore, it is usually safest to do a simple deduplication using the first range (prim_range). For example:
[0156] Relationship: Cohabitation: Basic(First Name: First Range: City Name Street:?: Second Range): Remove Duplicates(First Range)
[0157] RELATIONSHIP:COHABIT:BASIS(PRIM_NAME:PRIM_RANGE:CITY_NAME:ST:?:SEC_RANGE):DEDUP(PRIM_RANGE)
[0158] It should be noted that DEDUP performs intelligent deduplication, it will actually look for the strongest link for a given value of a deduplication field. In the example given above, if links are found with and without sec_range, it will use the link containing the sec_range value. If deduplication is omitted, it defaults to the entire basis. SALT also checks to make sure that all DEDUP elements come from the BASIS definition.
[0159] In a Scalable Automatic Linking Technology (SALT) relationship, fields are scored the same as internal links, and by default a link is declared if the internal link match threshold is exceeded. This can be overridden using the THRESHOLD parameter in the relationship statement. Since addresses typically have a specificity of about 28, 2 or more are typically required to declare a relationship. If you wish to enforce that a certain number of distinct links need to be found between clusters before a relationship is declared, use the MULTIPLE (n) option. Here n is the minimum number of links that must be found. A link occurs between any two entity identifiers that share a common value on a given basis.
[0160] Sometimes you want to be able to use additional information in the records to support a claim of a link, outside of the stated base, without requiring it to be part of the stated base. For example, it may not be significant that two identifiers share an address, but a relationship is more likely if the identifiers also share a Social Security Number (SSN) or share a last name. These supporting fields can be declared using the SCORE parameter in the RELATIONSHIP statement and its associated field list. For example:
[0161] Relationship: Cohabitation: Base(First Name: First Range: City Name Street:?: Second Range): De-duplicate(First Range): Score(Social Security Number: Last Name)
[0162] RELATIONSHIP:COHABIT:BASIS(PRIM_NAME:PRIM_RANGE:CITY_NAME:ST:?:SEC_RANGE):DEDUP(PRIMRANGE):SCORE(SSN:LNAME)
[0163] The fields in the scoring list are independent and will be used most heavily across all links participating in a relationship (if multiple relationship statements have the same field in the scoring list, the highest score for that field will be used for all relationship statements).
[0164] One of the problems with relations is getting them to run in a reasonable amount of time. Relations are a classic n-squared process, n-squared being correct for both time and disk space. Scalable Automatic Linking Technology (SALT) has done a lot of work on optimizing this process, which results in lightweight self-relationships being used in the generated ECL, but disk space utilization is still very efficient. The SPLIT(n) parameter in the relation statement allows the core relation to be split into multiple parts, each of which is persistent. This has the benefit of breaking very long relations into n parts (allowing for additional time periods), but also reduces disk consumption by a factor of n (assuming the final links are fairly sparse). In terms of performance, it should be noted that if n is high enough so that the output of each relation does not spill to disk, then the relation evaluation process will have significantly faster performance. The following example shows the use of the SPLIT(n) parameter:
[0165] Relationship: Cohabitation: Base(first name: first range: city name street:?: second range): Split(16): Score(last name: social security number): De-duplicate(first range)
[0166] RELATIONSHIP:COHABIT:BASIS(PRIM_NAME:PRIM_RANGE:CITY_NAME:ST:?:SEC_RANGE):SPLIT(16):SCORE(LNAME:SSN):DEDUP(PRIMRANGE)
[0167] Multiple RELATIONSHIP statements can be defined in one specification file. Each RELATIONSHIP is calculated independently. The output for the specified relations will be in the following format:
[0168] CombinationRecord: = RECORD
[0169] unsigned6ID1(unsigned 6 ID1);
[0170] unsigned6ID2(unsigned 6 ID2);
[0171] unsigned2Basis_score; / / Score allocated to the basisrelationship unsigned2Dedup_Val; / / Hashwill be stored here to dedup unsigned2Cnt; / / Number of different basis matches shared between / / clusters
[0172] unsignedl LNAME_score; / / Independent score for the LNAME field
[0173] unsignedl SSN_score (unsigned social security number score); / / Independent score for the SSN field END (independent score for the social security number field end);
[0174] ID1 and ID2 are the linked entity IDs. Basis_score is the cumulative score of all (de-duplicated) base matches that form a relationship. The Dedup_Val field is used for calculations and can be ignored. Cnt is the number of different base matches used in the relationship. The score for each score field is then derived, which has been maximized across all base matches in the relationship. The total score used to determine whether the match threshold is exceeded is Basis score + Lname score + Social Security number (SSN) score (Basisscore + Lnamescore + SSN score).
[0175] Any fields that are part of a concept definition (base or score) will be weighted appropriately when used in the base or score.
[0176] Relationships and dates
[0177] Tracks the date that relationship information was collected; this is provided you have a RECORDDATE field in your SPC. Any relationship can have a track list. TRACK is intended to provide date information about the relationship - it does not change the relationship calculation in any way (except perhaps slowing it down a bit). For example:
[0178] Relationship: Cohabitation: Base(first name: first range: city name: street: ?: second range): score(last name: social security number): deduplication(second range): threshold(35): tracking(first seen date: last seen date)
[0179] RELATIONSHIP:COHABIT:BASIS(PRIM_NAME:PRIM_RANGE:CITY_NAME:ST:?:SEC_RANGE):SCORE(LNAME:SSN):DEDUP(PRIM_RANGE):THRESHOLD(35):TRACK(dt_first_seen:dt_last_seen)
[0180] Relationship: Common Name: Base(First Name: Last Name): Multiple(2): Tracking(First Seen Date: Last Seen Date)
[0181] RELATIONSHIP:CONAME:BASIS(FNAME:LNAME):MULTIPLE(2):TRACK(dt_first_seen:dt_last_seen)
[0182] The system will then track the "first" date - the first date that both elements in the relationship have a fact on. For the "last" date, the last date that both elements in the relationship share that fact on. If a relationship is based on multiple coincidences, it will look for the earliest shared fact for the "first" date and the latest shared fact for the last date. This is a slightly odd combination of min and max, but semantically it works. For example:
[0183] From January 1, 1980 to January 1, 1990, A lived at 123 Main Street
[0184] From January 1, 1985 to January 1, 1987, B lived at 123 Main Street
[0185] From 19900101 to 20000101, A lived on 456 Street
[0186] From January 1, 1987 to January 1, 2010, B lived on 456 Street
[0187] The system first calculates the 'factual coincidence date':
[0188] From 19850101 to 19870101, A & B shared 123 Main Street (the shorter range when both were there)
[0189] From 19900101 to 200000101, A & B shared 456 Ave (the shorter range when both were there)
[0190] The earliest shared fact date is now 19850101 and the latest shared fact date is 20000101. These dates represent an upper bound when the relationship begins and a lower bound when it ends.
[0191] Note: If the shared facts do not overlap in time, the "first" date may be after the "last" date.
[0192] It is also possible to enforce a certain degree of date overlap within a single relationship. This is done by defining a range field on the two record dates. For example:
[0193] Field: First seen date: Record date (first seen, year, month): 5,0
[0194] FIELD:dt_first_seen:RECORDDATE(FIRST,YYYYMM):5,0
[0195] Field: Last seen date: Record date (last, year, month): 5,0
[0196] FIELD:dt_last_seen:RECORDDATE(LAST,YYYYMM):5,0
[0197] Range field: Report Date: Range (First seen date: Last seen date): 5,0
[0198] RANGEFIELD:ReportedDate:RANGE(dt_first_seen,dt_last_seen):5,0
[0199] Both record dates can be tracked if desired (or not). Report Date can now be used as a (fair) "regular" field in the base or scoring portion of a relationship. For example:
[0200] Relationship: Cohabitation: Base(first name: first range: city name: street: ?: second range): Score(social security number: last name: report date): Deduplication(first range): Threshold(35): Tracking(first seen date: last seen date)
[0201] RELATIONSHIP:COHABIT:BASIS(PRIM_NAME:PRIM_RANGE:CITY_NAME:ST:?:SEC_RANGE):SCORE(SSN:LNAME:REPORTEDDATE):DEDUP(PRIM_RANGE):THRESHOLD(35):TRACK(dt_first_seen:dt_last_seen)
[0202] Relationship: Common Name: Base (First Name: Last Name:?: Report Date): Multiple (2): Tracking (First Seen Date)
[0203] RELATIONSHIP:CONAME:BASIS(FNAME:LNAME:?:REPORTEDDATE):MULTIPLE(2):TRACK(dt_first_seen)
[0204] Relationship: Shared Social Security Number: Base(Social Security Number: Report Date): Score(Last Name): Threshold(35): Follow-up(Last Seen Date)
[0205] RELATIONSHIP:COSSN:BASIS(SSN:REPORTEDDATE):SCORE(LNAME):THRESHOLD(35):TRACK(dtlastseen)
[0206] Notice:
[0207] 1. In the rating, use global specificity with overlapping dates (same as for internal links)
[0208] 2. In the optional part of the basis, the dates may overlap or a record may have no date
[0209] 3. In the fixed part of the basis, there must be overlapping dates. Note: The range field cannot be the only (or first) in the basis
[0210] Combine relations with different bases into a single relation
[0211] It is also possible to define a relationship based on other relationships. This is done using a relationship list. All relationships in the list must be normal relationships, not grandparents. This definition uses the second form of the RELATIONSHIP statement. Consider the following example, which combines the Cohabitation relationship with the Shared Social Security Number relationship to form an association:
[0212] Relationship: Cohabitation: Base(first name: first range: city name: street: ?: second range): Split(16): Score(last name: social security number): Deduplication(first range): Threshold(30)
[0213] Relationship: Shared Social Security Number: Base(Social Security Number): Score(Surname): Deduplication(Social Security Number): Multiple(2) Relationship: Related: Cohabitation: Shared Social Security Number
[0214] RELATIONSHIP:COHABIT:BASIS(PRIM_NAME:PRIM_RANGE:CITY_NAME:ST:?:SEC_RANGE):SPLIT(16):SCORE(LNAME:SSN):DEDUP(PRIM_RANGE):THRESHOLD(30)
[0215] RELATIONSHIP:COSSN:BASIS(SSN):SCORE(LNAME):DEDUP(SSN):MULTIPLE(2)
[0216] RELATIONSHIP:ASSOC:COHABIT:COSSN
[0217] There are two normal relationships, one that ties the other two together into a unified score. It should be noted that since the normal relationships will ultimately be tied together by a correlation, you may want to weaken them using THRESHOLD statements, as shown in the example. Thus, a THRESHOLD of (30) on COHABIT may be sufficient to capture a very rare single address (40 is a typical person threshold in the LexisNexis people data). For shared SSNs, the MULTIPLE (2) parameter forces matching SSN pairs for the relationship; this is a more efficient approach because single SSN matches are already captured by the SCORE on COHABIT.
[0218] In principle, ASSOC then simply combines the two base relations and adds the scores; it is therefore very efficient and uses the stored results of the two normal relations. However, there is a problem between the sub-relations: it is easy for a lot of duplicate calculations to occur, because:
[0219] a) Both relations have the same field in the SCORE attribute (in this case, both have the last name), or
[0220] b) A scored attribute of one relationship may be part of the basis of another relationship (in this case, a cohabiting scored SSN is part of the basis of a shared SSN)
[0221] Scalable AutoLink Technology (SALT) automatically selects this and:
[0222] a) When multiple scores conflict, only the best score is selected
[0223] b) Assign the score from one relation to the underlying field of the other relation, then do the MAX assignment across those relations. (This is how single SSN matches are chosen even though it is prohibited for common SSNs).
[0224] The MULTIPLE and THRESHOLD parameters (or default values) are then applied to the child relationships. Thus, MULTIPLE (3) would require 3 different data matches to occur between cohabiting and shared social security numbers.
[0225] Many of the details of the subrelationship are preserved in the output generated for the composite relation. The format of the ASSOC example is:
[0226] shared ASSOCCRec:=RECORD (shared ASSOCCRec:=record)
[0227] unsigned6ID1;unsigned6ID2;(unsigned6ID1;unsigned6ID2)
[0228] unsigned2 Total_Score:=0; (unsigned 2 total score:=0)
[0229] unsigned2 Total Cnt 0; (unsigned 2 total Cnt 0)
[0230] unsigned2 COHABIT_score:=0; (unsigned2 cohabitation score:=0)
[0231] unsignedl COHABIT_cnt:=0; (unsigned1 cohabitation cnt:=0)
[0232] unsigned2 COSSN_score:=0; (unsigned2 common social security number score:=0)
[0233] unsignedl COSSN_cnt:=0; (unsigned2 common social security number cnt:=0)
[0234] unsignedl LNAME_score:=0; (unsigned1 last name score:=0)
[0235] / / Score for SSN will be rolled into COSSN as it is part of basis
[0236] END
[0237] Scalable Automatic Linking Technology (SALT) also builds separate keys for all relationships that are included as part of another relationship. In addition, a relationship service is established for each relationship that is not a child of another. This service takes a unique ID (UID) and a depth and will generate summary data for each relationship in a multi-layer deep tree. The treewalking aspect of this service is located in the relationship_links module. The tree is traversed so that each level of the tree is associated to a branch by the strongest link. At each level, IDI is the node being moved and ID2 is the node being moved to. The original node appears as a level 0 item in ID2.
[0238] For example, using the Sample_Input_Fileinternal AF Relationship_SPC specification file from the Scalable Automatic Linking Technology (SALT) sample module for internal linking produces the following results:
[0239]
[0240]
[0241] Result 1 is the file generated by the COFEIN relationship. This file contains the following results:
[0242] bdid1 bdid2 Basic score Remove duplicate val cnt Company Name Score 1 37178850 292680203 22 54286 2 11 2 292680203 37178850 22 54286 2 11
[0243] Result 2 is the file generated by the top-level relation named ASSOC. It contains the following results:
[0244]
Claims
1. A computer-implemented method for identifying at least one synthetic identity from a population of entity representations in an electronic database, wherein: Each entity representation includes a plurality of records, each record includes a plurality of fields, at least a first field of the plurality of fields can indicate the association of similar entity representations in the group, the method is performed by at least one processor configured to access the electronic database, the method includes: Select a first entity representation from the entity representation group; for the selected first entity representation: identifying one or more network chains formed by one or more associations between the first entity representation, the second entity representation, and zero or more intermediate entity representations; determining, in at least one of the identified one or more network chains, a number of associations for each association level of a predetermined level; Based on the number of associations at each association level, a geometric growth rate of the one or more associations between the first entity representation, the second entity representation, and the zero or more intermediate entity representations is determined, wherein the geometric growth rate r associated with a given association separation d can be determined by: Among them, x d is the correlation number at a given separation, x o is the initial association number; Based on the determined geometric growth rate being below a threshold, determining an indication that one or more of the first entity representation, the second entity representation, and the zero or more intermediate entity representations are composite identities; and Entity representation information of one or more entities in the group indicated as composite identities is output.
2. The method according to claim 1, characterized in that: Determining the number of associations for each association level includes: determining a shortest network link among the identified one or more network links; and The number of associations for each association level in the shortest network chain is determined.
3. The method according to claim 1, characterized in that: An indication that one or more of the first entity representation, the second entity representation, and the zero or more intermediate entity representations are composite identities is also determined based on an establishment date of the one or more associations.
4. The method according to claim 1, characterized in that: The one or more associations include one or more of a relative association, a business partner association, a social network association, a shared asset association, and a historical address association.
5. The method according to claim 1, characterized in that: The one or more associations are determined using a system map, wherein the system map includes information related to one or more public records, the public records including one or more of housing records, vehicle records, marriage records, divorce records, hospital records, death records, court records, property records, employment records, business records, address records, tax records, corporate records, incarceration records, and utility records, wherein the utility records include one or more of utility associations, disconnections, and associated service addresses.
6. The method according to claim 1, characterized in that: The one or more associations are based at least in part on distances between geographic areas proximate a plurality of locations associated with one or more activities common to the entity representations.
7. The method according to claim 1, characterized in that: The indication is also based on a difference between the determined geometric growth rates.
8. The method according to claim 1, characterized in that: The entity representation population includes 10 million or more entities.
9. The method according to claim 1, characterized in that: The predetermined level is 5 or less.
10. The method according to claim 1, characterized in that: For degrees d greater than 4, r less than 1.5 indicates a composite identity.
11. A computer-implemented system for identifying at least one synthetic identity among a population of entity representations in an electronic database, comprising: at least one memory for storing data and computer executable instructions; and at least one processor configured to access the at least one memory and further configured to execute the computer-executable instructions to identify one or more synthetic identities in a population of entity representations in an electronic database, Each entity representation includes a plurality of records, each record includes a plurality of fields, at least a first field of the plurality of fields can indicate an association of similar entity representations in the group, and the one or more synthetic identities are identified in the following manner: Select a first entity representation from the entity representation group, and for the selected first entity representation: identifying one or more network chains formed by one or more associations between the first entity representation, the second entity representation, and zero or more intermediate entity representations; determining, in at least one of the identified one or more network chains, a number of associations for each association level of a predetermined level; Based on the number of associations at each association level, a geometric growth rate of the one or more associations between the first entity representation, the second entity representation, and the zero or more intermediate entity representations is determined, wherein the geometric growth rate r associated with a given association separation d can be determined by: Among them, x d is the correlation number at a given separation, x o is the initial association number; Based on the determined geometric growth rate being below a threshold, determining an indication that one or more of the first entity representation, the second entity representation, and the zero or more intermediate entity representations are composite identities; and Entity representation information of one or more entities in the group indicated as synthetic identities is output.
12. The system according to claim 11, characterized in that: Determining the number of associations for each association level includes: determining a shortest network link among the identified one or more network links; and The number of associations for each association level of the shortest network chain is determined.
13. The system according to claim 11, characterized in that: An indication that one or more of the first entity representation, the second entity representation, and the zero or more intermediate entity representations are composite identities is determined also based on an establishment date of the one or more associations.
14. The system according to claim 11, characterized in that: The one or more associations are based at least in part on distances between geographic areas proximate to a plurality of locations associated with one or more activities common to the entities representing.
15. The system according to claim 11, characterized in that: For degrees d greater than 4, r less than 1.5 indicates a composite identity.
