Heuristic customer clustering
View Patent ↗Methods and apparatus are disclosed regarding an e-commerce system that places customers into a plurality of clusters and tailors services provided to a customer based on the cluster in which the customer is placed. In one embodiment, the e-commerce system defines the clusters based on purchase history data for customers having sufficient purchase history data. The e-commerce system then places customers without sufficient purchase history data into one of the defined clusters based on demographic data for the customer and demographic data for the customers in the cluster.
1. A computer-implemented method, comprising:
collecting purchase history data for a plurality of customers;
collecting demographic data for the plurality of customers;
splitting the plurality of customers into a first customer group comprising customers who have purchase transactions, within a time period of interest, greater than a predetermined number of purchase transactions and a second customer group comprising customers who have purchase transactions, within the time period of interest, not greater than the predetermined number of purchase transactions;
generating, based on the purchase history data of the first customer group, a matrix that relates a customer to purchased products to obtain a transaction space for the first customer group;
applying principal component analysis to the matrix to obtain a reduced matrix having a reduced transaction space for the first customer group;
generating, in accordance with a clustering large applications (CLARA) algorithm, a plurality of sample sets of the reduced matrix;
applying a partitioning around medoids (PAM) clustering algorithm to each of the plurality of sample sets to obtain a plurality of medoid sets comprising a medoid set per sample set;
clustering customers of the first customer group into a plurality of clusters based upon a medoid set of the plurality of medoid sets;
placing each customer of the second customer group into a cluster of the plurality of clusters that comprises a plurality of customers from the first customer group based on the collected demographic data for the customer and the collected demographic data for the plurality of customers from the first customer group placed in the respective cluster; and
tailoring services provided to a customer based on the cluster in which the customer resides.
2. The computer-implemented method of claim 1 , wherein the tailoring comprises providing product recommendations based on the cluster in which the customer resides.
3. The computer-implemented method of claim 1 , wherein the tailoring comprises providing product promotions based on the cluster in which the customer resides.
4. The computer-implemented method of claim 1 , wherein the tailoring comprises providing coupons based on the cluster in which the customer resides.
5. The computer-implemented method of claim 1 , wherein the tailoring comprises providing coupons based on the cluster in which the customer resides.
6. The computer-implemented method of claim 1 , further comprising validating the plurality of clusters prior to the placing each customer of the second customer group into a cluster.
7. The computer-implemented method of claim 1 , further comprising recalibrating a quantity of dimensions of the reduced transaction space until cluster validation determines the plurality of clusters are a valid clustering of the first customer group.
8. The computer-implemented method of claim 1 , further comprising recalibrating a quantity of clusters used by the clustering until cluster validation determines the plurality of clusters are a valid clustering of the first customer group.
9. A non-transitory computer-readable medium, comprising a plurality of instructions, that in response to being executed, result in a computing device:
collecting purchase history data for a plurality of customers;
collecting demographic data for the plurality of customers;
splitting the plurality of customers into a first customer group comprising customers who have purchase transaction, within a time period of interest, greater than a predetermined number of purchase transaction and a second customer group comprising customers who have purchase transactions, within the time period of interest, not greater than the predetermined number of purchase transaction;
generating, based on the purchase history data of the first customer group, a matrix that relates a customer to purchased products to obtain a transaction space for the first customer group;
applying principal component analysis to the matrix to obtain a reduced matrix having a reduced transaction space for the first customer group;
generating, in accordance with a clustering large applications (CLARA) algorithm, a plurality of sample sets of the reduced matrix;
applying a partitioning around medoids (PAM) clustering algorithm to each of the plurality of sample sets to obtain a plurality of medoid sets comprising a medoid set per sample set;
clustering customers of the first customer group into a plurality of clusters based upon a medoid set of the plurality of medoid sets;
placing each customer of the second customer group into a cluster of the plurality of clusters that comprises a plurality of customers from the first customer group based on the collected demographic data for the customer and the collected demographic data for the plurality of customers from the first customer group placed in the respective cluster; and
tailoring services provided to a customer based on the cluster in which the customer resides.
10. The non-transitory computer-readable medium of claim 9 , further comprising instructions that result in the computing device validating the plurality of clusters prior to the placing each customer of the second customer group into a cluster.
11. The non-transitory computer-readable medium of claim 9 , further comprising instructions that result in the computing device recalibrating a quantity of dimensions of the reduced transaction space until cluster validation determines the plurality of clusters are a valid clustering of the first customer group.
12. The non-transitory computer-readable medium of claim 9 , further comprising instructions that result in the computing device recalibrating a quantity of clusters used by the clustering until cluster validation determines the plurality of clusters are a valid clustering of the first customer group.
13. A computing device, comprising
an electronic database comprising demographic data and purchase history data for a plurality of customers; and
a processor configured to:
split the plurality of customers into a first customer group comprising customers who have purchase transactions, within a time period of interest, greater than a predetermined number of purchase transactions and a second customer group comprising customers who have purchase transactions, within the time period of interest, not greater than the predetermined number of purchase transactions;
generate, based on the purchase history data of the first customer group, a matrix that relates a customer to purchased products to obtain a transaction space for the first customer group;
apply principal component analysis to the matrix to obtain a reduced matrix having a reduced transaction space for the first customer group;
generate, in accordance with a clustering large applications (CLARA) algorithm, a plurality of sample sets of the reduced matrix;
apply a partitioning around medoids (PAM) clustering algorithm to each of the plurality of sample sets to obtain a plurality of medoid sets comprising a medoid set per sample set;
cluster customers of the first customer group into a plurality of clusters based upon a medoid set of the plurality of medoid sets;
place each customer of the second customer group into a cluster of the plurality of clusters that comprises a plurality of customers from the first customer group based on the collected demographic data for the customer and the collected demographic data for the plurality of customers from the first customer group placed in the respective cluster; and
tailor services provided to a customer based on the cluster in which the customer resides.
14. The computing device of claim 13 , wherein the processor is further configured to recalibrate a quantity of dimensions of the reduced transaction space and a quantity of clusters for the plurality of clusters until cluster validation determines the plurality of clusters are a valid clustering of the first customer group.
15. The computing device of claim 13 , wherein the processor is further configured to:
calculate, for each medoid set, a sum of distances to a nearest medoid of the medoid set; and
select the medoid set that has a smallest sum of distances for said clustering customers of the first customer group.
16. The computer-implemented method of claim 1 , further comprising:
calculating, for each medoid set, a sum of distances to a nearest medoid of the medoid set; and
selecting the medoid set that has a smallest sum of distances for said clustering customers of the first customer group.
17. The non-transitory computer-readable medium of claim 9 , further comprising instructions that result in the computing device:
calculating, for each medoid set, a sum of distances to a nearest medoid of the medoid set; and
selecting the medoid set that has a smallest sum of distances for said clustering customers of the first customer group.