Systems and methods for management of multi-perspective customer segments
Systems and methods for management of multi-perspective customer segments. This invention relates to customer management, and more particularly to management of customer segments, wherein the customer segments can be multi-perspective. Embodiments herein disclose methods and systems for generating multiple sets of clusterings of customers based on at least one customer attribute, and combining and interactively visualizing the clusterings, to derive insights.
1. A system for generating multiple sets of clusters that are connected through a hierarchical structure from a dataset pertaining to user behaviour, said system comprising
a processor;
a memory storing operational instructions which, when executed by said processor, cause the processor to:
define, using a clustering module, a plurality of views by grouping said dataset based on multiple segmentation attributes of user behaviour;
perform outlier treatment on said plurality of views to filter out outlier data points that form part of the dataset;
based on an automatic determination of optimal number of clusters for each of said plurality of views using non-parametric clustering, generate said optimal number of clusters in said each of said plurality of views; and
in response to generation of said optimal number of clusters in said each of said plurality of views, combine, using a consolidation module, clusters from said plurality of views to generate said hierarchical structure that allows interactive discovery of a pattern of interest through recursive analysis of the clusters across the plurality of views in different feature subsets, wherein the clusters from said plurality of views are represented across different levels of the hierarchical structure in a manner such that if parent nodes of said hierarchical structure are represented by clusters of a first view of said plurality of views, child nodes of each said parent node are formed of clusters having users that are common with users of said respective parent node to avoid re-clustering of data within each of said plurality of views while identifying said pattern of interest, and wherein a first child node is associated with at least two parent nodes.
2. The system of claim 1 , wherein said clustering module defines said plurality of views using feature subspace selection.
3. The system of claim 1 , wherein said outlier treatment of the said plurality of views comprises:
computing first and third quartile values for each of said one or more segmentation attributes;
performing, for each data point of said dataset, an outlier test based on Inter-Quartile Range (IQR) for each of said one or more segmentation attributes such that if data point value for respective segmentation attribute is below first quartile value or above third quartile value of said respective segmentation attribute, said data point is determined to be outlier.
4. The system of claim 1 , wherein said optimal number of clusters in said each of said plurality of views are generated by starting with a single cluster whose mean equals global centroid, and iteratively forming a new cluster whenever a data point of said dataset of said view is farther than lambda away from every existing cluster centroid till convergence.