Abstract
In this paper we propose an in-depth analysis of a method, called the flow network method, which associates with any network a complete and quasi-transitive binary relation on its vertices. Such a method, originally proposed by Gvozdik (Abstracts of the VI-th Moscow conference of young scientists on cybernetics and computing. Scientific Council on Cybernetics of RAS, Moscow, p 56, 1987), is based on the concept of maximum flow. Given a competition involving two or more teams, the flow network method can be used to build a relation on the set of teams which establishes, for every ordered pair of teams, if the first one did at least as good as the second one in the competition. Such a relation naturally induces procedures for ranking teams and selecting the best k teams of a competition. Those procedures are proved to satisfy many desirable properties.
Similar content being viewed by others
Notes
The National Football League (NFL) and the NCAA Division I Football Bowl Subdivision are examples of competitions where there are teams which never play each other on the field.
We describe a competition by a table in which every row has the shape
meaning that the matches involving the teams x and y were \(m+n\) and that x won m times and y won n times.
We stress that Belkin and Gvozdik (1989) mainly focus on the problem of building rankings.
A similar interpretation appears in Patel (2015).
Due to the identification between competitions and networks, we are going to freely use the terminology of competitions for networks too.
Note that in many applications networks capacities are allowed to be nonnegative real numbers and networks are identified with their adjacency matrix.
The Copeland method is sometimes called net flow method (Bouyssou 1992).
We emphasize that some of the properties considered by González-Díaz et al. (2014) are satisfied by the flow network method as described in course of the paper.
The flow network method fulfils the three main properties stated in that paper. Property I (opponent strength) has been already discussed; Property II (incentive to win) is the content of Proposition 11; Property III (sequence of matches) follows from the very definition of the method.
Note that the (normalized) invariant and the fair-bets methods are also known as the (normalized) long path and the Markov methods, respectively.
A tournaments is a complete and asymmetric digraph. Tournaments can be identified with networks whose capacities are 0 or 1 and such that the sum of the capacities of any pair of opposite arcs is 1. They are used to represent round robin-competitions.
A balanced network is a network whose sum of the capacities of any pair of opposite arcs is constant. Balanced networks represent competitions where any pair of teams confront each other the same number of time.
The equality between the flow and the Borda network methods on balanced networks is stated, without proof, in Belkin and Gvozdik (1989).
Complete and quasi-acyclic relations always admit linear refinements (Proposition 27).
That fact can be easily checked considering the network in Example 56.
Given a relation R on a nonempty finite set V, a k-maximum set of R is a subset W of size k of V having the property that, for every \(x\in W\) and \(y\not \in W\), \((x,y)\in R\). Complete and quasi-acyclic relations always admit k-maximum sets (Proposition 30).
Weak tournaments and partial tournaments can be naturally identified with networks whose arcs always have capacities that are 0 or 1 and such that the sum of the capacities of each pair of opposite arcs is at least 1 and at most 1 respectively.
Those authors use the term anonymity instead of neutrality.
For a more general approach, see also the main result in Hartmann and Schneider (1993).
We thank László Lovász for useful advices about this proposition.
That fact seems to be known in the literature even though we could not find a precise reference. Thus, for completeness, we provide a proof.
Note that, on the network \(N_{\textsf {D}}\), the Kemeny network rule and the ranked pair network rule are equal to \(\{L, \,\textsc {c}\succ \textsc {b}\succ \textsc {a},\, \textsc {a}\succ \textsc {c}\succ \textsc {b}\}\), while \(\mathfrak {F}_{\diamond }(N_{\textsf {D}})=\{L\}\).
Actually, this definition is less general than the one in the paper of Dutta and Laslier. Nevertheless, it is sufficient for our purposes.
References
Aziz H, Brill M, Fischer F, Harrenstein P, Lang J, Seeding HG (2015) Possible and necessary winners of partial tournaments. J Artif Intell Res 54:493–534
Bang-Jensen J, Gutin G (2008) Digraphs theory. Algorithms and applications. Springer, New York
Belkin AR, Gvozdik AA (1989) The flow model for ranking objects. In: Dorofeyuk A, Litvak B, Tjurin Yu (eds) Problems of cybernetics, decision making and analysis of expert information, Moscow, pp 109–117 (in Russian)
Bouyssou D (1992) Ranking methods based on valued preference relations: a characterization of the net flow method. Eur J Oper Res 60:61–67
Brandt F, Brill M, Harrenstein P (2016) Extending tournament solutions. In: Proceedings of the twenty-eighth AAAI conference on artificial intelligence
Bubboloni D, Gori M (2015) Symmetric majority rules. Math Soc Sci 76:73–86
Bubboloni D, Gori M (2016a) On the reversal bias of the minimax social choice correspondence. Math Soc Sci 81:53–61
Bubboloni D, Gori M (2016b) Resolute refinements of social choices correspondences. Math Soc Sci 84:37–49
Csató L (2017) Some impossibilities of ranking in generalized tournaments. arXiv:1701.06539v5
De Donder P, Le Breton M, Truchon M (2000) Choosing from a weighted tournament. Math Soc Sci 40:85–109
Dutta B, Laslier J-F (1999) Comparison functions and choice correspondences. Soc Choice Welf 16:513–532
Elkind E, Faliszewski P, Skowron P, Slinko A (2017) Properties of multiwinner voting rules. Soc Choice Welf 48:599–632
Gomory RE, Hu TC (1961) Multi-terminal network flows. J Soc Ind Appl Math 9(4):551–570
González-Díaz J, Hendrickx R, Lohmann E (2014) Paired comparisons analysis: an axiomatic approach of ranking methods. Soc Choice Welf 42:139–169
Gvozdik AA (1987) Flow interpretation of the tournament model for ranking. Abstracts of the VI-th Moscow conference of young scientists on cybernetics and computing. Scientific Council on Cybernetics of RAS, Moscow, p 56 (in Russian)
Hartmann M, Schneider MH (1993) Flow symmetry and algebraic flows. ZOR Methods Models Oper Res 38:261–267
Kitti M (2016) Axioms for centrality scoring with principal eigenverctors. Soc Choice Welf 46:639–653
Langville AN, Meyer CD (2012) Who is #1? The science of rating and ranking. Princeton University Press, Princeton
Laslier J-F (1997) Tournament solutions and majority voting. Studies in Economic Theory, vol 7. Springer, Berlin
Lovász L (1973) Connectivity in digraphs. J Combin Theory (B) 15:174–177
Palacios-Huerta I, Volij O (2004) The measurement of intellectual influence. Econometrica 72:963–977
Patel J (2015) College football rankings: maximum flow model. Industrial Engineering Undergraduate Honors Thesis, Paper 35
Peris JE, Subiza B (1999) Condorcet choice correspondences for weak tournaments. Soc Choice Welf 16:217–231
Saari DG (1994) Geometry of voting. In: Studies in economic theory, vol 3. Springer, Berlin
Saari DG, Barney S (2003) Consequences of reversing preferences. Math Intell 25:17–31
Schulze M (2011) A new monotonic, clone-independent, reversal symmetric, and Condorcet-consistent single-winner election method. Soc Choice Welf 36:267–303
Slutzki G, Volij O (2005) Ranking participants in generalized tournaments. Int J Game Theory 33:255–270
Szpilrajn E (1930) Sur l’extension de l’ordre partiel. Fundam Math 16:386–389
van den Brink R, Gilles RP (2009) The outflow ranking method for weighted directed graphs. Eur J Oper Res 193:484–491
Vaziri B, Dabadghao S, Yih Y, Morin TL (2017) Properties of sports ranking methods. J Oper Res Soc. https://doi.org/10.1057/s41274-017-0266-8
Author information
Authors and Affiliations
Corresponding author
Additional information
We wish to thank an anonymous referee for letting us know the existence of the papers by Gvozdik (1987) and Belkin and Gvozdik (1989) (in Russian), where the flow network method was first formulated. We also thank Andrey Sarychev for translating the mentioned papers. Daniela Bubboloni was partially supported by GNSAGA of INdAM.
Rights and permissions
About this article
Cite this article
Bubboloni, D., Gori, M. The flow network method. Soc Choice Welf 51, 621–656 (2018). https://doi.org/10.1007/s00355-018-1131-7
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00355-018-1131-7