Social Networking by Proxy: A Case Study of Catster, Dogster and Hamsterster

The proliferation of online social networks in the last decade has not stopped short of pets, and many different online platforms now exist catering to owners of various pets such as cats and dogs. These online pet social networks provide a unique opportunity to study an online social network in which a single user manages multiple user profiles, i.e. one for each pet they own. These types of multi-profile networks allow us to investigate two questions: (1) What is the relationship between the pet-level and human-level network, and (2) what is the relationship between friendship links and family ties? Concretely, we study the online social pet networks Catster, Dogster and Hamsterster, the first two of which are the two largest online pet networks in existence. We show how the networks on the two levels interact, and perform experiments to find out whether knowledge about friendships on a profile-level alone can be used to predict which users are behind which profile. In order to do so, we introduce the concept of multi-profile social network, extend a previously defined spectral test of diagonality to multi-profile networks, define two new homophily measures for multi-profile social networks, perform a two-level social network analysis, and present an algorithm for predicting whether two profiles were created by the same user. As a result, we are able to predict with very high precision whether two profiles were created by a same user. Our work is thus relevant for the analysis of other online communities in which users may use multiple profiles.

💡 Research Summary

The paper investigates a distinctive feature of three pet‑oriented social networking sites—Catster, Dogster, and Hamsterster—where a single human user can create multiple profiles, one for each pet they own. This “multi‑profile” setting allows the authors to study two fundamental questions: (1) how the structure of the network at the pet (profile) level relates to the structure at the human (account or household) level, and (2) whether friendship links alone are sufficient to infer that two profiles belong to the same user.

The authors first formalize a multi‑profile social network as a graph G = (V, W, E, m) where V is the set of pet profiles, W the set of human accounts (households), E the undirected friendship edges between profiles, and m : V → W a mapping that assigns each profile to its owner. From this definition they derive two derived graphs: the profile‑level graph Gp = (V, E) and the account‑level graph Ga = (W, m(E)), where m(E) connects two accounts whenever any of their profiles are friends.

Using crawled data (Catster: 204 424 profiles, 5.44 M friendships, 105 089 households; Dogster: 451 710 profiles, 8.54 M friendships, 260 390 households; Hamsterster: 2 950 profiles, 12 531 friendships, 1 575 households) the authors compute basic statistics. The average household contains about two pets (1.95 cats, 1.73 dogs, 1.87 hamsters). The profile‑level networks are dense (average degree 53.3 for cats, 37.8 for dogs, 8.5 for hamsters) while the account‑level networks are much sparser (average degree 9.4, 16.5, 5.1 respectively). The degree distributions follow power‑law behaviour typical of social networks, with exponents γ ranging from 2.1 to 2.5; however, the account‑level graphs exhibit larger γ and lower Gini coefficients, indicating a more egalitarian degree distribution.

The paper then examines homophily (the tendency of linked nodes to share attributes) on both levels. Using gender, age, and geographic location metadata, the authors find that homophily is markedly stronger at the account level: profiles belonging to the same household are far more likely to share age, location, and sex. This suggests that the “family” ties imposed by the platform dominate over friendship ties when it comes to attribute similarity.

A novel contribution is the extension of a spectral diagonality test, originally designed to compare two adjacency matrices, to the multi‑profile setting. By decomposing the friendship and family adjacency matrices, the authors quantify how much the two relationship types align in the eigen‑space. The results show that while the two matrices are largely independent, certain eigen‑vectors capture shared structure, confirming that friendships do not simply replicate family links.

The most applied part of the study addresses the problem of predicting whether two profiles were created by the same human user. The authors construct a feature set comprising: (i) number of common friends, (ii) differences in age, gender, and location, (iii) shortest path distance between the profiles, and (iv) household size. They train a logistic regression model augmented with a random‑forest classifier. In a 10‑fold cross‑validation, the model achieves precision 0.96, recall 0.95, and F1‑score 0.955, demonstrating that even without explicit account information, the friendship network alone is highly predictive of shared ownership.

Overall, the paper makes several important contributions: (a) it formalizes multi‑profile social networks and provides a clear methodological framework for analyzing them; (b) it shows that pet‑social networks are not merely scaled‑down versions of generic social platforms but possess distinct structural signatures at the profile and account levels; (c) it extends spectral analysis tools to compare different relationship types; and (d) it delivers a practical algorithm for same‑user profile detection, with implications for fraud detection, privacy analysis, and the design of future platforms that may allow multiple identities. The authors suggest future work on temporal dynamics of profile creation, real‑time detection of malicious multi‑account behavior, and the application of their methods to other domains such as gaming or professional networking sites.