[Paper Review] eBay users form stable groups of common interest

Joerg Reichardt, Stefan Bornholdt|arXiv (Cornell University)|Mar 16, 2005

Consumer Market Behavior and Pricing3 references17 citations

TL;DR

This paper proposes a network-based community detection method to identify stable, overlapping groups of eBay users with shared interests, using only bidding competition data without relying on article similarity measures or taxonomic hierarchies. The key finding is that user interest profiles remain highly stable over time, indicating enduring consumer milieus shaped by personal expertise and social reinforcement.

ABSTRACT

Market segmentation of an online auction site is studied by analyzing the users' bidding behavior. The distribution of user activity is investigated and a network of bidders connected by common interest in individual articles is constructed. The network's cluster structure corresponds to the main user groups according to common interest, exhibiting hierarchy and overlap. Key feature of the analysis is its independence of any similarity measure between the articles offered on eBay, as such a measure would only introduce bias in the analysis. Results are compared to null models based on random networks and clusters are validated and interpreted using the taxonomic classifications of eBay categories. We find clear-cut and coherent interest profiles for the bidders in each cluster. The interest profiles of bidder groups are compared to the classification of articles actually bought by these users during the time span 6-9 months after the initial grouping. The interest profiles discovered remain stable, indicating typical interest profiles in society. Our results show how network theory can be applied successfully to problems of market segmentation and sociological milieu studies with sparse, high dimensional data.

Motivation & Objective

• To study market segmentation on eBay by analyzing user bidding behavior without relying on predefined article similarity measures or taxonomic hierarchies.
• To detect stable, coherent user clusters based solely on competition in auctions, reflecting shared interests.
• To validate the stability of identified interest profiles over time using post-grouping purchase behavior.
• To demonstrate that network-based community detection can reveal sociological milieus in sparse, high-dimensional online transaction data.

Proposed method

• Constructed a network where nodes represent bidders and edges represent competition in at least one auction, capturing shared interest in articles.
• Applied a community detection algorithm to identify clusters of users with common bidding patterns, allowing for hierarchical and overlapping structures.
• Used random network null models to validate the non-random, non-trivial cluster structure in the bidding network.
• Interpreted clusters using eBay’s official product category taxonomy, without using it for data reduction or similarity computation.
• Tracked user activity over time by sampling clusters and analyzing their purchase behavior 6–9 months later to test profile stability.
• Calculated odds ratios for category-specific bidding and buying to quantify and compare interest profiles across clusters.

Experimental results

Research questions

• RQ1 Can user groups with common interests be reliably detected in online auction data without relying on article similarity measures?
• RQ2 Do the interest profiles of user clusters remain stable over time, despite the availability of diverse products on eBay?
• RQ3 How do the structural properties of the bidding network—such as hierarchy and overlap—reflect real-world consumer behavior?
• RQ4 To what extent do user preferences align with eBay’s official product categorization, and can this taxonomy be used to interpret latent interest clusters?

Key findings

• 85% of eBay users could be classified into a small number of well-separated, large clusters, each with a distinct and coherent set of main interests.
• The interest profiles of users in each cluster remained remarkably stable over a 6–9 month period, with only minor shifts in secondary interests.
• Users in cluster 1 shifted from primarily bidding on movies to a stronger focus on music, while cluster 9 (technology-affine users) showed broader interest diversity.
• Collectors (cluster 7) and toy model builders (cluster 4) exhibited minimal change in their profiles, suggesting long-term, stable preferences.
• The stability of interest profiles suggests that users tend to bid only on articles they have prior experience with, reducing risk from unfamiliar categories.
• Online recommender systems may reinforce existing interest profiles, contributing to the long-term persistence of user clusters.