CUSTOMER SEGMENTATION VIA CONSENSUS CLUSTERING
The customer segmentation system includes a basic partition constructor to generate basic partitions of customers in an original feature space. Further, a partition space transformer in the customer segmentation system transforms the original feature space to an augmented partition space based on membership information of the customers into the basic partitions. Subsequently, a consensus clustering builder in the customer segmentation system determines consensus-based partitions of the customers in the augmented partition space. As such, robust and high-quality partitions for customer segmentation are achieved in the customer segmentation system.
1 . One or more non-transient computer storage media comprising computer-implemented instructions that, when used by one or more computing devices, cause the one or more computing devices to:
receive information of a plurality of customers associated with an original feature space;
generate a first plurality of partitions of the plurality of customers based on a first plurality of sequential partitioning stages operated using the original feature space, wherein a first partitioning stage generates more partitions than a second partitioning stage in the first plurality of sequential partitioning stages;
build an augmented partition space based on membership information of the plurality of customers in the first plurality of partitions; and
determine a second plurality of partitions of the plurality of customers based on a second plurality of sequential partitioning stages operated using the augmented partition space.
2 . The one or more computer storage media of claim 1 , the instructions further cause the one or more computing devices to:
randomly select, from the plurality of customers, a first predetermined number of customers as a set of candidates for a present stage of the first plurality of partitioning stages;
generate a second predetermined number of partitions of the plurality of customers in the original feature space for each candidate of the set of candidates; and
determine an objective function value associated with each candidate in the present stage.
3 . The one or more computer storage media of claim 2 , wherein the first predetermined number is in a range of 40 to 80.
4 . The one or more computer storage media of claim 2 , wherein the objective function value indicates a distance measure of the plurality of customers from a respective partition center.
5 . The one or more computer storage media of claim 2 , the instructions further cause the one or more computing devices to:
add, to a set of partition centers, a candidate having a minimum objective function value in the present stage;
generate partitions of the plurality of customers for the present stage based on the set of partition centers; and
update the set of partition centers based on the partitions of the plurality of customers for the present stage.
6 . The one or more computer storage media of claim 1 , the instructions further cause the one or more computing devices to:
construct a binary matrix to represent the augmented partition space with binary elements in the binary matrix corresponding to the membership information of the plurality of customers in the first plurality of partitions.
7 . The one or more computer storage media of claim 1 , the instructions further cause the one or more computing devices to:
concatenate the membership information of the plurality of customers in the first plurality of partitions;
construct a binary matrix to represent the augmented partition space with binary elements in the binary matrix corresponding to the concatenated membership information.
8 . The one or more computer storage media of claim 1 , wherein the first plurality of sequential partitioning stages has more stages than the second plurality of sequential partitioning stages.
9 . The one or more computer storage media of claim 1 , the instructions further cause the one or more computing devices to:
generate the first plurality of partitions of the plurality of customers in the original feature space based on a greedy K-means clustering method.
10 . A computer-implemented method, comprising:
receiving information of a plurality of customers associated with an original feature space and a cluster number;
generating a first plurality of partitions of the plurality of customers through a first plurality of sequential partitioning stages operated in the original feature space;
transforming the original feature space to an augmented partition space based on membership information of the plurality of customers in the first plurality of partitions; and
determining a second plurality of partitions of the plurality of customers in the augmented partition space based on the cluster number and a second plurality of sequential partitioning stages operated using the augmented partition space.
11 . The method of claim 10 , further comprising:
randomly selecting a predetermined number of customers from the plurality of customers as a set of candidates for a present stage of the first plurality of sequential partitioning stages; and
generating respective partitions of the plurality of customers in the original feature space for each candidate of the set of candidates.
12 . The method of claim 11 , further comprising:
measuring distances from the plurality of customers to their respective partition centers associated with a candidate; and
determining an objective function value associated with the candidate based on the measured distances.
13 . The method of claim 12 , further comprising:
adding a candidate having a minimum objective function value in the present stage to a set of partition centers determined at a prior stage;
generating partitions for the present stage based on the set of partition centers; and
updating the set of partition centers based on the partitions for the present stage.
14 . The method of claim 10 , further comprising:
constructing a data structure to represent the augmented partition space with elements in the data structure corresponding to membership information of the plurality of customers in the first plurality of partitions.
15 . The method of claim 10 , wherein the first plurality of sequential partitioning stages has more stages than the second plurality of sequential partitioning stages.
16 . A system, comprising:
means for generating a first plurality of partitions of a plurality of customers in a first plurality of sequential partitioning stages operated using an original feature space;
means for transforming the original feature space into an augmented partition space based on membership information of the plurality of customers in the first plurality of partitions; and
means for determining a second plurality of partitions of the plurality of customers in a second plurality of partitioning stages operated using the augmented partition space.
17 . The system of claim 16 , the system further comprising:
means for randomly selecting a predetermined number of customers from the plurality of customers as a set of candidates for a present stage of the first plurality of partitioning stages;
means for generating respective partitions for the plurality of customers in the original feature space for each candidate of the set of candidates; and
means for determining an objective function value associated with each candidate in the present stage based on respective partitions associated with each candidate.
18 . The system of claim 17 , the system further comprising:
means for adding a candidate having a minimum objective function value in the present stage to a set of partition centers generated from a prior stage of the first plurality of partitioning stages;
means for generating partitions for the present stage based on the set of partition centers; and
means for updating the set of partition centers based on the partitions for the present stage.
19 . The system of claim 16 , wherein the original feature space comprises indications of customer features related to visitor activity information, traffic pattern information, referral data information, advertising campaign information, visitor retention information, or product data information.
20 . The system of claim 16 , wherein the augmented partition space comprises a binary matrix with binary elements representing the membership information of the plurality of customers in the first plurality of partitions.