Back to results
US 12,586,102 B2Granted
Heuristic clustering
AssigneeTransform Sr Brands LLC
InventorsBurcu Aydin, Michael Tamir
Life of this patent
Estimating the term…
Abstract
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.
Claims21
A method, the method comprising:
- splitting a plurality of records into a first group and a second group;
- generating a matrix associated with data of the first group, wherein the matrix comprises a Customer-Item (CI) matrix that is standardized column-wise using bin-quantiles standardization (BQS) to produce a standardized transaction space with values between 0 and 1;
- applying principal component analysis to the standardized transaction space to obtain a reduced matrix for the first group;
- generating, in accordance with a clustering large applications (CLARA) procedure, a plurality of samples of the first group according to the reduced matrix;
- partitioning each of the plurality of samples of the first group to reduce dissimilarities;
- applying a partitioning-around-medoids (PAM) algorithm using Euclidean distance to each sample;
- computing for each medoid set a sum of distances of all points to a nearest medoid;
- selecting the medoid set having the smallest sum of distances;
- clustering the first group into a plurality of clusters according to the selected medoid set;
- and
- placing a record of the second group into a particular cluster, of the plurality of clusters, according to demographic data of the record and demographic data of the particular cluster.
8 more elements
A non-transitory computer readable medium, comprising a plurality of instructions, that in response to being executed, result in a computing device:
- splitting a plurality of records into a first group and a second group;
- generating a matrix associated with data of the first group, wherein the matrix comprises a Customer-Item (CI) matrix that is standardized column-wise using bin-quantiles standardization (BQS) to produce a standardized transaction space with values between 0 and 1;
- applying principal component analysis to the standardized transaction space to obtain a reduced matrix for the first group;
- generating, in accordance with a clustering large applications (CLARA) procedure, a plurality of samples of the first group according to the reduced matrix;
- partitioning each of the plurality of samples of the first group to reduce dissimilarities;
- applying a partitioning-around-medoids (PAM) algorithm using Euclidean distance to each sample;
- computing for each medoid set a sum of distances of all points to a nearest medoid;
- selecting the medoid set having the smallest sum of distances;
- clustering the first group into a plurality of clusters according to the selected medoid set;
- and
- placing a record of the second group into a particular cluster, of the plurality of clusters, according to demographic data of the record and demographic data of the particular cluster.
8 more elements
A system, the system comprising:
- one or more processors configured to:
- split a plurality of records into a first group and a second group;
- generate a matrix associated with data of the first group, wherein the matrix comprises a Customer-Item (CI) matrix that is standardized column-wise using bin-quantiles standardization (BQS) to produce a standardized transaction space with values between 0 and 1;
- apply principal component analysis to the standardized transaction space to obtain a reduced matrix for the first group;
- generate, in accordance with a clustering large applications (CLARA) procedure, a plurality of samples of the first group according to the reduced matrix;
- partition each of the plurality of samples of the first group to reduce dissimilarities;
- apply a partitioning-around-medoids (PAM) algorithm using Euclidean distance to each sample;
- compute for each medoid set a sum of distances of all points to a nearest medoid;
- select the medoid set having the smallest sum of distances;
- cluster the first group into a plurality of clusters according to the selected medoid set;
- and
- place a record of the second group into a particular cluster, of the plurality of clusters, according to demographic data of the record and demographic data of the particular cluster.
9 more elements
Dependencies are inferred from the claim text.