Metric-based identity resolution
AssigneeQuantavalue L.L.C.
InventorsYang Liu, Jun Li
Life of this patent
Estimating the term…
Abstract
Digital records are clustered and assigned unique individual identity across a plurality of data sources. Each record contains one or more tagged common properties, which belongs to an underlying identifiable entity. A special parameterized metric distance is calculated for all pairs of records. With adjustment of the metric parameters, the records are clustered based on a predetermined distance threshold. Based on the clustered records, the underlying identity associated with the records can be resolved.
Claims6
1-11. (canceled)
A computer-implemented method for resolving underlying identity for digital data records across a plurality of data sources, the method comprising:
- building a system that assigns association with individual identities for a plurality of structured or unstructured data records;
- parsing the data based on a set of tagged common properties;
- using the parsed common properties to derive a calculated distance between pairs of data records, wherein the method of deriving a calculated distance comprises defining a property-wise metric distance δ, and calculating a record-wise metric distance D(a, b), where (a, b) is a pair of records and each record has n common properties, by transforming property-wise metric distances δ(ai, bi), where (ai, bi) is a pair of properties consisting of the i-th property for a and b respectively, with scaling factors si and translation factors ti, according to a formula
- D
-
- (
- a
- ,
- b
- )
- =
- ∏
- i
- =
- 1
- n
-
- [
- s
- i
-
- δ
-
- (
- a
- i
- ,
- b
- i
- )
- +
- t
- i
- ]
- ;
- generating a linkage graph by establishing linkage between pairs of data records with a calculated distance below a given threshold, filtering the calculated distances and adjusting the scaling and translation factors in the distance calculation accordingly;
- and
- clustering the data records based on the linkage graph and assigning each clustered data records a unique identity.
A system for resolving underlying identity for digital data records across a plurality of data sources, the system comprising:
- a processor;
- a memory device having memory accessible by the processor;
- and
- processor-executable instructions stored in the memory of the memory device, the instructions executable to instruct the system to:
- building a system that assigns association with individual identities for a plurality of structured or unstructured data records;
- parsing the data based on a set of tagged common properties;
- using the parsed common properties to derive a calculated distance between pairs of data records, wherein the method of deriving a calculated distance comprises defining a property-wise metric distance δ, and calculating a record-wise metric distance D(a, b), where (a, b) is a pair of records and each record has n common properties, by transforming property-wise metric distances δ(ai, bi), where (ai, bi) is a pair of properties consisting of the i-th property for a and b respectively, with scaling factors si and translation factors ti, according to a formula
- D
-
- (
- a
- ,
- b
- )
- =
- ∏
- i
- =
- 1
- n
-
- [
- s
- i
-
- δ
-
- (
- a
- i
- ,
- b
- i
- )
- +
- t
- i
- ]
- ;
- generating a linkage graph by establishing linkage between pairs of data records with a calculated distance below a given threshold, filtering the calculated distances and adjusting the scaling and translation factors in the distance calculation accordingly;
- and
- clustering the data records based on the linkage graph and assigning each clustered data records a unique identity.
A non-transitory computer-readable media comprising a plurality of instructions which, executable by a processor, cause the processor to:
- building a system that assigns association with individual identities for a plurality of structured or unstructured data records;
- parsing the data based on a set of tagged common properties;
- using the parsed common properties to derive a calculated distance between pairs of data records, wherein the method of deriving a calculated distance comprises defining a property-wise metric distance δ, and calculating a record-wise metric distance D(a, b), where (a, b) is a pair of records and each record has n common properties, by transforming property-wise metric distances δ(ai, bi), where (ai, bi) is a pair of properties consisting of the i-th property for a and b respectively, with scaling factors si and translation factors ti, according to a formula
- D
-
- (
- a
- ,
- b
- )
- =
- ∏
- i
- =
- 1
- n
-
- [
- s
- i
-
- δ
-
- (
- a
- i
- ,
- b
- i
- )
- +
- t
- i
- ]
- ;
- generating a linkage graph by establishing linkage between pairs of data records with a calculated distance below a given threshold, filtering the calculated distances and adjusting the scaling and translation factors in the distance calculation accordingly;
- and
- clustering the data records based on the linkage graph and assigning each clustered data records a unique identity.
Dependencies are inferred from the claim text.