Measuring Customer Similarity and Identifying Cross-Selling Products by Community Detection

Area of Science:

• Data Science
• Marketing Analytics
• Computational Social Science

Background:

• Customer segmentation is key for targeted marketing, but traditional methods struggle with large, sparse datasets and generalizable insights.
• Existing approaches like market basket analysis can yield overly broad associations, limiting actionable strategies.
• Challenges include skewed data, computational complexity, and achieving meaningful, interpretable customer groups.

Purpose of the Study:

• To develop an efficient and effective method for product affinity segmentation.
• To improve the accuracy and actionability of customer segmentation for cross-selling and loyalty programs.
• To address the limitations of traditional clustering and market basket analysis in big data contexts.

Main Methods:

• Proposed a novel approach using community detection within a customer-product bipartite graph.
• Applied the Louvain algorithm to partition customers into groups based on product purchase similarity.
• Conducted a case study with data from a large U.S. retailer to validate the method.

Main Results:

• The Louvain algorithm generated interpretable customer clusters with distinct product purchase patterns.
• Customer and product characteristics within clusters were statistically significant and driven by purchase data.
• The proposed method outperformed the conventional Recency, Frequency, Monetary (RFM) model in recommendation response rates.

Conclusions:

• The community detection approach offers a powerful alternative for product affinity segmentation in big data.
• This method enhances understanding of customer behavior and improves product recommendation effectiveness.
• The approach effectively addresses computational complexity and data sparsity issues inherent in large-scale retail data.