Fraud detection systems and methods
Methods and systems, including computer programs encoded on computer storage media, for detecting fraudulent or suspicious transactions, e.g., in a business-to-business context. One of the methods includes: receiving, at an ingestion engine, raw data from a plurality of heterogeneous data sources; using the ingestion engine to produce structured object data, the structured object data derived from the raw data; applying a classification service to produce a partial fingerprint from the structured object data wherein the partial fingerprint classifies the structured object data; applying a fingerprinting process to the partial fingerprint to produce a full fingerprint for the structured object data; applying an enrichment engine to add data to the full fingerprint to produce an enriched fingerprint for the structured object data; and applying a verdict engine to determine a verdict on the object based at least in part on the enriched fingerprint for the structured object data.
1 . A method comprising:
receiving, at an ingestion engine, raw data from a plurality of heterogeneous data sources, the raw data including transaction data recorded at different dates;
using the ingestion engine to produce structured object data for an object, the structured object data derived from the raw data;
applying a classification service to produce a partial fingerprint from the structured object data wherein the partial fingerprint comprises a plurality of attributes and wherein the classification service classifies the structured object data based at least in part on the plurality of attributes;
applying a fingerprinting process to the partial fingerprint to produce a full fingerprint for the structured object data, the full fingerprint comprising financial attributes;
applying an enrichment engine to add data to the full fingerprint to produce an enriched fingerprint for the structured object data, wherein applying the enrichment engine comprises applying a network engine to the full fingerprint to determine a network score for the enriched fingerprint, the network score based at least in part on trust network data wherein the trust network data comprises anonymized fingerprint data for a plurality of objects, the anonymized fingerprint data comprising data from users; and
applying a verdict engine to determine a verdict on the object based at least in part on financial attributes of the enriched fingerprint for the structured object data.
2 . The method of claim 1 , wherein applying the verdict engine comprises accessing a customer baseline as part of determining the verdict on the object.
3 . The method of claim 2 , wherein applying the classification service comprises applying an initial filter to filter out data that is not relevant to constructing a fingerprint for structured object data.
4 . The method of claim 3 , wherein applying the classification service to produce the partial fingerprint from the structured object data comprises applying a machine learning process to the structured object data.
5 . The method of claim 4 , wherein applying the fingerprinting process comprises applying a natural language processing engine to characterize the structured object data.
6 . The method of claim 5 , wherein applying the fingerprinting process comprises applying a layout engine to the structured object data to determine a layout of the structured object data.
7 . The method of claim 6 , applying the machine learning process comprises taking the structured object data as at least one input to provide at least a portion of the full fingerprint.
8 . The method of claim 7 , wherein applying the fingerprinting process comprises providing a version of the full fingerprint to a trust network engine.
9 . The method of claim 3 , wherein applying the fingerprinting process comprises applying the fingerprinting process to structured object data for a plurality of objects describing interactions with a specified entity and wherein applying the fingerprinting process comprises
applying an analysis engine to structured object data for each of the plurality of objects to produce a plurality of fingerprints, each of the plurality of fingerprints corresponding to one of the plurality of objects and wherein the method further comprises:
applying a correlation engine to the plurality of fingerprints to produce a correlated fingerprint object;
repeating the applying the fingerprinting process and applying the correlation engine steps over time to produce a plurality correlated fingerprint objects; and
applying a baseline model engine to the plurality of correlated fingerprint objects to produce a baseline for an entity.
10 . The method of claim 1 , wherein the anonymized fingerprint data is non-hash data.
11 . The method of claim 10 , wherein the financial attributes comprise at least one transaction attribute, at least one vendor attribute or at least one payment attribute or a combination of at least two such attributes.
12 . The method of claim 1 , wherein applying the enrichment engine comprises obtaining enrichment data from a trust network engine.
13 . A system comprising:
one or more computers having one or more processors and one or more storage devices on which are stored instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
receiving, at an ingestion engine, raw data from a plurality of heterogeneous data sources, the raw data including transaction data recorded at different dates;
using the ingestion engine to produce structured object data for an object, the structured object data derived from the raw data;
applying a classification service to produce a partial fingerprint from the structured object data wherein the partial fingerprint comprises a plurality of attributes and wherein the classification service classifies the structured object data based at least in part on the plurality of attributes;
applying a fingerprinting process to the partial fingerprint to produce a full fingerprint for the structured object data, the full fingerprint comprising financial attributes;
applying an enrichment engine to add data to the full fingerprint to produce an enriched fingerprint for the structured object data, wherein applying the enrichment engine comprises applying a network engine to the full fingerprint to determine a network score for the enriched fingerprint, the network score based at least in part on trust network data wherein the trust network data comprises anonymized fingerprint data for a plurality of objects, the anonymized fingerprint data comprising data from users; and
applying a verdict engine to determine a verdict on the object based at least in part on financial attributes of the enriched fingerprint for the structured object data.
14 . The system of claim 13 , wherein applying the verdict engine comprises accessing a customer baseline as part of determining a verdict on the object.
15 . The system of claim 14 , wherein applying the classification service comprises applying an initial filter to filter out data that is not relevant to constructing a fingerprint for structured object data.
16 . The system of claim 15 , wherein applying the classification service to produce a partial fingerprint from the structured object data comprises applying a machine learning process to the structured object data.
17 . The system of claim 16 , wherein applying the fingerprinting process comprises applying a natural language processing engine to characterize the structured object data.
18 . The system of claim 17 , wherein applying the fingerprinting process comprises applying a layout engine to the structured object data to determine a layout of the structured object data.
19 . The system of claim 15 , wherein applying the fingerprinting process comprises applying the fingerprinting process to structured object data for a plurality of objects describing interactions with a specified entity and wherein applying the fingerprinting process comprises
applying an analysis engine to structured object data for each of the plurality of objects to produce a plurality of fingerprints, each of the plurality of fingerprints corresponding to one of the plurality of objects and wherein the method further comprises:
applying a correlation engine to the plurality of fingerprints to produce a correlated fingerprint object;
repeating the applying a fingerprinting process and applying a correlation engine steps over time to product a plurality correlated fingerprint objects; and
applying a baseline model engine to the plurality of correlated fingerprint objects to produce a baseline for an entity.
20 . A method comprising:
receiving, at an ingestion engine, raw data from a plurality of heterogeneous data sources the raw data including transaction data recorded at different dates;
using the ingestion engine to produce structured object data for an object, the structured object data derived from the raw data;
applying a classification service to produce a partial fingerprint from the structured object data wherein the partial fingerprint classifies the structured object data, wherein applying the classification service comprises applying an initial filter to filter out data that is not relevant to constructing a fingerprint for structured object data;
applying a fingerprinting process to the partial fingerprint to produce a full fingerprint for the structured object data, wherein applying the fingerprinting process comprises applying a natural language processing engine to characterize the structured object data, applying a layout engine to the structured object data to determine a layout of the structured object data, applying a machine learning process that takes the structured object data as at least one input to provide at least a portion of the full fingerprint, the full fingerprint comprising financial attributes;
applying an enrichment engine to add data to the full fingerprint to produce an enriched fingerprint for the structured object data; and
applying a verdict engine to determine a verdict on the object based at least in part on financial attributes of the enriched fingerprint for the structured object data, wherein applying the verdict engine comprises accessing a customer baseline as part of determining a verdict on the object.