Clusterization methods for multi-variant e-commerce interfaces

Abstract

E-commerce is a very popular method that lets consumers to purchase goods and services. The ability to purchase items online has increased the need for effective recommendation systems. Such recommendations usually relete to products in which the customer may be interested. However, there are wider opportunities to tailor e-commerce to individual customer needs and behaviour. In this paper, the architecture of the e-commerce platform (named AIM 2 ), which allows providing a dedicated interface to selected user groups, is discussed. A key component of the platform is the module responsible for dividing customers into groups, using selected clustering methods. Each of the implemented methods can be parameterised to adapt the customer segmentation to a given e-commerce business owner's requirements. This article describes the results of an analysis of the impact of selected methods and parameters on clustering results. Moreover, it identifies key metrics that should be considered when selecting clustering conditions during the implementation of the platform. Finally, the main results of the pilot implementation of AIM 2 are presented to assess the effectiveness of the multi-variant user interface.

Functional Framework for Multivariant E-Commerce User Interfaces

“…In addition, BIRCH [54] and agglomerative clustering [55] may be promising approaches. Based on the previous research [8], it can be concluded that the agglomerative clustering method give better results from the point of view of the distribution of the clusters, but this may be impractical for the analysis of large datasets due to its computational complexity. During the experiments, the K-means model provided results that were not much worse than agglomerative clustering.…”

“…While there are various approaches to creating adaptive user interfaces, one of the most effective appears to be the use of artificial intelligence (AI)-based clustering to divide clients into groups and provide those groups with a dedicated interface [8]. AI clustering is a technique that groups similar data points (e.g., e-commerce customers) based on certain characteristics.…”

One Size Does Not Fit All: Multivariant User Interface Personalization in E-Commerce

“…[18]- [21] E-commerce layout personalization A form of personalization that results in the customization of a comprehensive UI design, taking into account content and layout.…”

“…When deciding on a clustering algorithm, there are three sets of factors to consider: computational complexity, overall clustering quality, and business context applicability. Among the clustering methods that have been analyzed for the application of multi-variant UI in ecommerce are DBSCAN (Density-Based Spatial Clustering of Applications with Noise), BIRCH (Balanced Iterative Re-ducing and Clustering using Hierarchies), GMM (Gaussian Mixture Model), Agglomerative Clustering [20], K-means and Spectral clustering [21]. The size of data sets containing information about ecommerce customer behavior forces careful selection of the clustering algorithm due to computational requirements.…”

“…The last but not least, the business critical factor influencing the choice of clustering algorithm should be the verification of its suitability to the business context, leading to the practical utility of the resulting clusters. In the case of a multivariant UI, customer groups should be as similar in size as possible, and the number of customers in a single cluster should not be less than the chosen threshold [21]. These requirements arise from the potential cost of designing dedicated UI variants, which is only worthwhile if the UI variant will be served to a sufficiently large number of users.…”