High-dimensional market subdivision method and device based on canonical correlation

Abstract

The invention discloses a canonical correlation-based high-dimensional market subdivision method and device, and relates to the technical field of market subdivision, and the method comprises the following steps: standardizing a plurality of groups of original variables, and mapping the standardized original variables to a high-dimensional feature space through a kernel function; constructing a variable inter-group covariance matrix, and solving typical variable pairs to maximize inter-group correlation; and performing clustering analysis by using typical variables to form a subdivided group and analyze the characteristics of the subdivided group. According to the canonical correlation analysis-based high-dimensional market subdivision method provided by the invention, kernel functions (such as a linear kernel and a Gaussian kernel) and grid search optimization parameters are combined, so that nonlinear relationships among consumer population characteristics, behavior modes, psychological demands and consumption scenes are effectively captured, and the limitation of a traditional linear model is overcome. Through multi-variable collaborative analysis and dynamic weighted clustering, the multi-source data fusion capability and the result interpretability are improved, and the method is suitable for complex market subdivision scenes.

Images