Partitioning signed social networks☆
Introduction
Some social relations are signed in the sense of having ties between actors that can be positive or negative. Examples include like/dislike and respect/disrespect for people. Analyses of the network structure of signed relations have to accommodate the additional information contained in the negative part of the signed ties. Structural balance theory, one variant of ‘consistency’ theories, has its origins in the work of Heider (1946). Cartwright and Harary (1956) then formalized Heider’s formulation and provided the foundation for discerning the overall structure of a network of signed ties. This structure for a ‘balanced’ network is given by the first structure theorem described below. A generalization proposed by Davis (1967) leads to the second structure theorem. However, for most observed signed structures for social groups, exact structural balance does not hold. Even so, it is still necessary to delineate the structure of these groups at points in time. Based on the structure theorems, Doreian and Mrvar (1996) proposed an empirical method for establishing the partition structure(s) of a signed relation for groups that is (are) as close to exact balance as is possible. Their method was subsumed within the rubric of generalized blockmodeling (Doreian et al., 2005). Even though these methods have been useful, the blockmodel structure discerned may not be appropriate for all signed networks for groups. We provide some illustrative examples and then broaden the types of blockmodel that can be specified and identified for signed networks within the generalized blockmodeling framework. We then apply this new blockmodel type to four real data sets and provide commentary on the new results.
Section snippets
Structural balance theory and partitions
In Cartwright and Harary’s (1956) generalization of Heider’s (1946) formulation, the difference between sentiment (social) relations and unit formation relations was ignored with attention confined, in effect, to signed social relations. Following Doreian et al. (2005), a binary signed network is an ordered pair, , where:
- (1)
is a digraph, without loops, having a set of units (vertices), , and a set of arcs, ; and
- (2)
is a sign function. The arcs with the sign p are positive
Blockmodels of signed social networks
The generalized blockmodeling approach is a direct method that analyzes the network data rather than some transformation of them. It requires a criterion function that is optimized by a relocation algorithm. See Doreian et al. (2005) for details of this approach to delineating network structure. The criterion function for signed networks is designed in terms of the structure theorems via a count of elements that are not consistent with an ideal k-balance partition. These inconsistencies take
Relaxing the structural balance blockmodel
When the signed network shown in Fig. 2 is partitioned according to structural balance for . With this implies five ties are inconsistent with structural balance. (For there are six optimal partitions, each with 4.5 as the value of the criterion function. From Theorem 3, the unique partition for is the best possible partition under structural balance.) The partition structure is given in Table 3 where the five inconsistent ties are bolded and italicized. The
The newcomb data (last week)
The Newcomb data (see Nordlie, 1958, Newcomb, 1961) are well known and have been analyzed many times. Originally, the data were reported with rankings made by 17 men of each other while they lived in a pseudo-dormatory at a university. Previously unknown to each other, these men provided sociometric data that were used to study the evolution of network ties over time. Doreian et al. (2005) report the shift over time in the overall structure leading to a blockmodel of the signed relation at the
Discussion
Partitioning based on structural balance has proven useful for delineating the structure of signed social networks. While most observed signed social networks have not been balanced, in the sense of having a line index of balance that is non-zero, the delineated structures have been interpretable. For the few signed networks where data have been available over time, a tendency towards balance has been observed, notably in the Sampson (1968) data (Doreian and Mrvar, 1996) where the measure of
Conclusion
We have proposed a generalization of structural balance where the notion of positive and negative blocks has been retained but with the modification that they can appear anywhere in the blockmodel of a signed network. Assuming that structural processes leave traces found in the network structure, as delineated by blocks, it is now possible to identify structural features additional to those implied by the operation of structural balance processes. Differential popularity, differential receipt
References (21)
- et al.
Constructing blockmodels: how and why
Journal of Mathematical Psychology
(1978) - et al.
An algorithm for clustering relational data with applications to social network analysis and comparison to multidimensional scaling
Journal of Mathematical Psychology
(1975) A multiple indicator approach to blockmodeling signed relations
Social Networks
(2008)- et al.
A partitioning approach to structural balance
Social Networks
(1996) Comparison of methods for positional analysis: structural and general equivalences
Social Networks
(1988)- et al.
Structures of connectivity
Social Networks
(1984) Personality and Interpersonal Behavior
(1970)- et al.
Correlation clustering
Machine Learning
(2004) - et al.
Pajek—program for large network analysis
Connections
(1998) - et al.
Structural balance: a generalization of Heider’s theory
Psychological Review
(1956)
Cited by (186)
A review of heuristics and metaheuristics for community detection in complex networks: Current usage, emerging development and future directions
2021, Swarm and Evolutionary ComputationA signed network analysis-based Consensus Reaching Process in Group Decision Making
2021, Applied Soft ComputingA general model for how attributes can reduce polarization in social groups
2023, Network ScienceTowards a computational network theory of social groups
2022, Behavioral and Brain SciencesSHEEP, a Signed Hamiltonian Eigenvector Embedding for Proximity
2024, Communications PhysicsA new approach to finding overlapping community structure in signed networks based on Neutrosophic theory
2024, Journal of Complex Networks
- ☆
Support from the Slovenian Research Agency under the Grant, Contract number 1000-07-780005, is greatly appreciated.