Category-Based Routing in Social Networks: Membership Dimension and the Small-World Phenomenon (Short)

A classic experiment by Milgram shows that individuals can route messages along short paths in social networks, given only simple categorical information about recipients (such as “he is a prominent lawyer in Boston” or “she is a Freshman sociology major at Harvard”). That is, these networks have very short paths between pairs of nodes (the so-called small-world phenomenon); moreover, participants are able to route messages along these paths even though each person is only aware of a small part of the network topology. Some sociologists conjecture that participants in such scenarios use a greedy routing strategy in which they forward messages to acquaintances that have more categories in common with the recipient than they do, and similar strategies have recently been proposed for routing messages in dynamic ad-hoc networks of mobile devices. In this paper, we introduce a network property called membership dimension, which characterizes the cognitive load required to maintain relationships between participants and categories in a social network. We show that any connected network has a system of categories that will support greedy routing, but that these categories can be made to have small membership dimension if and only if the underlying network exhibits the small-world phenomenon.

💡 Research Summary

The paper investigates how messages can be routed through social networks using only categorical information about the target, a phenomenon famously illustrated by Milgram’s “six degrees of separation” experiments. The authors formalize this intuition by modeling a social network as an undirected graph G = (U,E) and a collection S ⊂ 2^U of overlapping “categories” (groups) that each vertex belongs to. For a vertex u, cat(u) denotes the set of categories containing u.

A key contribution is the definition of membership dimension (memdim), the maximum number of categories any single vertex belongs to. This quantity captures the cognitive load a participant must maintain: the smaller the memdim, the fewer group memberships a person needs to remember in order to make routing decisions.

The routing rule is purely greedy: given a current node s and a destination t, the distance function is d(s,t) = |cat(t) \ cat(s)|, i.e., the number of target categories not shared by s. Node s forwards the message to a neighbor v with strictly smaller distance, d(v,t) < d(s,t). The rule requires only local knowledge of a neighbor’s categories and the target’s categories, matching the limited information available to humans.

Two structural properties of the pair (G,S) are identified as necessary (and, for trees, sufficient) for the greedy algorithm to succeed for every source‑destination pair:

1. Internally connected – each category induces a connected subgraph of G.
2. Shattered – for any distinct s and t, there exists a neighbor u of s and a category C that contains both u and t but not s.

The shattered property guarantees that from any node there is at least one neighbor that shares more categories with the target, enabling progress. In trees, internal connectivity together with shattering ensures that the greedy rule never gets stuck (Lemma 2). In general graphs these two conditions are not sufficient; a counter‑example is provided where no neighbor reduces the distance.

The authors then explore the relationship between memdim and the graph’s diameter (the longest shortest‑path distance). They prove a lower bound (Lemma 3): any category system that enables greedy routing must satisfy memdim ≥ diam(G). Intuitively, a message must traverse at least diam(G) steps, and each step reduces the distance by at most one, so the target must belong to at least diam(G) distinct categories.

Conversely, they show an upper bound: for any connected graph on n vertices, one can construct a category system with memdim = O((diam(G)+log n)²) that guarantees greedy routing (Theorem 1). The construction proceeds by first taking a spanning tree of G, then assigning to each vertex a hierarchy of “prefix” and “suffix” categories along the tree paths (similar to interval labeling). Additional O(log n) auxiliary categories are added to handle non‑tree edges, ensuring the shattered property while keeping the total number of categories per vertex bounded by a quadratic function of the diameter plus a logarithmic term.

For special cases the bounds are tight. On a simple path (line) graph, a straightforward construction using left‑interval and right‑interval categories yields memdim = diam(G) (Lemma 4). For a star (height‑1 tree) a comparable construction also achieves the lower bound.

The practical implication is that small‑world networks, which empirically have diameters that grow at most logarithmically with the number of nodes, also admit category systems with polylogarithmic membership dimension. Hence, humans can plausibly maintain only a modest number of group affiliations while still being able to forward messages greedily toward distant targets.

The paper situates its contribution among related work on geometric greedy routing, virtual coordinate systems, and previous sociological studies of categorical routing. While geometric approaches rely on physical or virtual coordinates, the present work treats categories as high‑dimensional “coordinates” where the sparsity (low memdim) is the analogue of succinctness.

In summary, the authors provide a rigorous theoretical framework linking three concepts: (i) the existence of short paths (small‑world property), (ii) the cognitive feasibility of maintaining a limited set of overlapping groups (membership dimension), and (iii) the success of a simple greedy routing algorithm based solely on categorical overlap. Their results both explain why Milgram‑type experiments succeed and offer design principles for decentralized routing protocols in ad‑hoc or delay‑tolerant networks where categorical or interest‑based information is more readily available than precise geographic location.