Over the past few years, the problem of influence and spread in networks has been subject to intense study (Ballester et al., 2006; Bharathi et al., 2007; Galeotti and Goyal, 2009; Kempe et al., 2003, 2005; Chasparis and Shamma, 2010; Vetta, 2002). Furthermore, modeling and analysis of the spread of new strategies and behaviors via local coordination games has been an ongoing field of research (Ellison, 1993; Kandori et al., 1993; Harsanyi and Selten, 1988; Young, 1993, 2001, 2002; Montanari and Saberi, 2010; Kleinberg, 2007). For example, in López-Pintado (2006) the authors show that the contagion of an action in a random network depends on the distribution of the connectivities. In Amini et al. (2009), authors provide an upper bound on the proportion of agents adopting a new product assuming a threshold model for product adoption. Montanari and Saberi (2010) studies the diffusion of innovation in social networks based on the dynamics of coordination games and shows that innovation spreads much more slowly on well-connected network structures dominated by long-range links than in low-dimensional ones dominated by geographic proximity, contrasting some earlier works on epidemic models (e.g., (Ganesh et al., 2005; Draief et al., 2006)).
A game theoretic model of competition and product adoption has been proposed in Goyal and Kearns (2012). The authors use the proposed model to come up with upper bounds on the price of anarchy and show how network structure may increase the gap between the initial budgets. Similarly, in Bimpikis et al. (2013), the authors propose a game theoretic model for competition between firms, where firms can target their marketing budgets toward attracting individual consumers embedded in a social network. They subsequently provide conditions under which it is optimal for the firms to asymmetrically target a subset of the individuals. As another relevant work, (Chasparis and Shamma, 2010) considers a dynamical model of preferences in a duopoly setting and characterize optimal policies for both finite and infinite time horizons, studying the effect of endogenous network influences as well as network uncertainties. The equilibria of network games with linear best response dynamics have been completely characterized in Bramoullé et al. (2014). Considering a monopoly setting, optimal pricing policies are derived in Candogan et al. (2012) assuming quadratic utility functions for the agents, which is a common theme in game-theoretic social network analysis as also previously used in Ballester et al. (2006); Corbo et al. (2007) for instance.
Despite the tremendous development made in the past decade (see e.g., (Seeman and Singer, 2013; Fazeli et al., 2017; Bharathi et al., 2007; Montanari and Saberi, 2010; Chasparis and Shamma, 2010)), influence maximization algorithms are typically devised under idealized assumptions regarding accessibility of individual consumers for targeted advertisement (e.g., via seeding) as well as the linearity of the cost function with respect to the size of the influence on individual’s consumption behavior. In practice, however, firms often have direct access only to a small subset of consumers. In addition, the extent to which firms can influence consumption behavior of consumers is limited, no matter how much they spend on seeding/advertisement. This can be more formally stated as diminishing returns on changes in individual consumption levels, which we model by assuming a convex seeding cost function in our work. It seems also quite compelling to investigate approaches that can support sparse seeding as (at least) near-optimal marketing strategies, in order to account for the limitations in directly accessing individuals especially in large populations.111By sparse seeding we mean only seeding a subset of consumers with an infinitesimal size compared to the size of the whole population.
In this paper, we study strategic competition between two firms seeking to maximize their product consumption in a network. The consumption of each product by each agent is the result of her myopic best response to the previous consumptions of her peers. A firm can thus improve its market share by targeting its advertising budget toward seeding key individuals in the network, whose consumption of the product can in turn incentivize their peers to consume more of the same product, subsequently affecting the consumption behavior of the individuals all over the network via inter-agent influences. We model the problem above as a fixed-sum game between the two firms, where each firm tries to maximize a utility function which is a discounted sum of its product consumption over time less a seeding cost which is appropriately chosen to account for the diminishing return in seeding/advertisement budget as well. We characterize the unique Nash equilibrium of the resulting duopoly game in terms of the network structure and the market price. The resultant seeding strategy typically prescribes seeding all the agents (in an amount proportional to their influence), which is rather an infeasible task. As a remedy, we propose studying the -equilibria of the game and derive conditions under which such equilibria can asymptotically lead to sparse seeding strategies in large populations.
The rest of this paper is organized as follows: in Section 2, we present basic mathematical notations. In Section 3, we introduce our model and update dynamics for agents by applying the myopic best response. We then study the game played between the firms and how they decide to seed key individuals in Section 4. Next, we define a near-Nash equilibrium concept as a relaxation of the standard Nash equilibrium with the aim of expanding the equilibrium set to include sparse seeding strategies in Section 5. In Section 6, we consider the case of a large population and characterize network structures for which a pair of sparse seeding strategies can be asymptotically realized. The results are illustrated via an example in Section 7. Finally, in Section 8, we conclude the paper.
2 Mathematical Notations
Throughout the paper, the discrete time index is denoted by . The sets of real (integer), positive real (integer), and strictly positive real (integer) numbers are represented by (), () and (), respectively. The set of natural numbers is denoted by . Bold letters, such as or
, stand for real-valued vectors. Capital letters, such asor , stand for real-valued matrices. We use to denote the -norm of vector . We denote the number of nonzero elements in vector by . The -by-identity matrix is denoted by . Also, we represent the -by- vector of ones by and the -by- and -by- matrices of zeros by and , respectively. The transpose of matrix is denoted by . For two matrices and , we denote the Hadamard product by and the Kronecker product by .
3 Spread Dynamics
We consider a social network consisting of a group of consumers (agents) denoted by . The relationship among agents is given by a weighted directed graph . The weighted adjacency matrix of is denoted by where its -th entry denoted by if otherwise . Link weight presents the strength of the influence of agent on .
We assume there are two competing firms producing product and , respectively. Let and denote the in-degree and out-degree of node , respectively. Assume and denote agent ’s consumption of product and at time , respectively. The initial consumption of agent from each product is determined by the effort made by each firm in seeding/marketing its product to agent at time , and are denoted by and . We refer to and as the control/seeding of firm and on agent , respectively. The relation between the controls and the states is motivated by (Chasparis and Shamma, 2010) on social networks. We consider a convex cost function for seeding to reflect the diminishing return of seeding/advertisement budget on changing the consumption behavior in the population. The cost reflects the monetary value required to increase the consumption of the agents in the population by amount . For the sake of simplicity, we develop our results for a quadratic cost function of the form .
The total utility of agent from taking action is given by
where . In the above equation denotes an action vector of all agents but agent . The first part of the utility () represents the normalized second order approximation of a concave self-utility function, as is commonly used in the literature. The fact that externality is weaker for consuming different products is captured by . Finally, the price represents the common market price for the products.
3.1 Myopic Best Response Dynamics
We assume agents repeatedly apply myopic best response to the consumption levels of their neighbors in the previous stage to update their consumption of each product, that is,
Consumption levels for agent at time thus find the following linear dynamics:
This results in the following closed-form update dynamics:
where and .
We assume to guarantee that and are non-negative for all feasible initial seedings.
In the next section, we show how firms can exploit the structure of the network to maximize their product consumption, and we then characterize the unique Nash equilibrium of the game played between these two firms.
We first recall the definition of the Katz-Bonacich centrality measure.
For a given attenuation factor less than the reciprocal of the absolute value of the largest eigenvalue of the adjacency matrix
less than the reciprocal of the absolute value of the largest eigenvalue of the adjacency matrix, the Katz-Bonacich centrality is given by
For a given subset , we define an indicator binary vector as follows:
We then use these definitions in the next section to characterize equilibrium strategies in a duopoly game between the firms.
4 Optimal Seeding Strategies
This section describes the game between two firms where each firm aims to maximize the consumption of its product over an infinite horizon. Each firm can invest in promoting its product by seeding some of the agents. Initial seeding could be viewed as free offers to promote the product in the social networks. We thus consider the problem of deriving optimal advertising policies for the spread of innovations/consumption in a network. In Chasparis and Shamma (2010), an analytical solution to the optimal advertising problem in the absence of a competing firm is provided, and it is shown that the solution can be related to previously introduced centrality measures in sociology.
We define the utility of each firm as the discounted sum of its product consumption over time minus the squared norm of its seeding, which are defined formally as follows:
A pair of seeding strategies is said to be a Nash equilibrium of the duopoly game described above if none of the players can improve her payoff by unilaterally deviating from her strategy. That is,
We assume that the absolute value of the largest eigenvalue of is less than or equal to where
In the following lemma, an equilibrium strategy for the duopoly game between the firms is characterized based on the node centrality measures.
Let us define the following Jacobian matrices for firm as follows:
Similarly, we define the following Jacobian matrices for firm :
Next, we write the update dynamics for and according to their definitions and update dynamics (4):
where , and . We can rewrite (17) in the following compact form
From this, it follows that
We then use the following property to simplify (18)
Let us define
With a simple calculation we get:
From (6), we get
The proof is a direct consequence of Lemma 1.
5 Sparse Seeding
In what follows, we first define a near-Nash equilibrium concept as a relaxation of the standard Nash equilibrium defined in Definition 3, with the aim of expanding the equilibrium set to include sparse seeding strategies (cf. (Daskalakis et al., 2006)). In what follows, we will make this statement formal.
Given , a pair of seeding strategies is said to be -equilibrium of the duopoly game described in Section 4 if none of the players can improve her payoff by an amount more than fraction of her current payoff, by unilaterally deviating from her strategy. That is,
Every Nash Equilibrium is equivalent to a -equilibrium where .
The next lemma provides a closed-form expression for the utility functions of the firms when there is not any seeding.
Assume , then the utility of firm is reduced to
where centrality is given by Definition 1.
Let us assume , then
where is given by (20). With a simple calculation we get:
Then, based on the definition of the utility, we have
6 Asymptotically Realizable Sparse Equilibria
In this section, we consider the case of a large population for which . We characterize network structures for which a pair of sparse seeding strategies can be realized as the limit of a sequence of -equilibria with . In what follows, we will make this statement formal.
We call a pair of -equilibrium seeding strategies asymptotically sparse-realizable (ASR) if and only if , and .222Given functions and , the asymptotic notations and mean and , respectively.
We begin our analysis by a lemma that presents a necessary and sufficient condition for a pair of strategies be asymptotically sparse-realizable in terms of the bi-product centrality (10).
A pair of strategies is asymptotically sparse-realizable if and only if
where is given by (10), , , , and .
We first start with the fact that if and only if . Therefore, the pair of -equilibrium seeding strategies is asymptotically sparse-realizable if and only if
Let us recall
where , and . It can be seen that
where in the last inequality, we use the fact that .
As the next result, we derive a necessary condition for the existence of asymptotically sparse-realizable equilibrium strategies.
Let , and suppose that
Then, there exists no pair of -equilibrium seeding strategies that is asymptotically sparse-realizable.
This result implies that networks with bounded out-degree (i.e., ) are not asymptotically sparse-realizable.
We conclude this section by stating a sufficient condition for the existence of asymptotically sparse-realizable equilibrium strategies.
The proof is a direct consequence of Lemma 3.
We illustrate our results using a core-periphery network structure, that is a network with few highly interconnected and many sparsely connected nodes.
Consider a network consisting of communities of size denoted by for . Consumption levels of the agents within each community are influenced by an agent called a role model that induces an externality effect of magnitude on each consumer. Consumption levels of role models themselves are each influenced by the consumption of a role model in another community whose structure is assumed to form a cycle. Let agent be the role model in community for . The corresponding entries of the adjacency matrix are given by
The case is depicted in Fig. 1.
Characterizing the equilibrium seeding strategies (both Nash equilibrium and -equilibria) requires finding centrality vectors
where . Let and be the corresponding centralities of a role model and a periphery consumer in , respectively. A periphery consumer does not induce externality on any other consumer, resulting in a centrality of . For a role model, on the other hand, from the definition of Katz-Bonacich centrality we can obtain
resulting in . Similarly, we can find and .
Applying Theorem 1, the symmetric Nash equilibrium strategy thus involves seeding a periphery consumer by , and every role model by an amount of
Keeping the number of communities fixed and shifting , we can use Lemma 3 to verify that seeding only the role models according to (7) is an asymptotically sparse-realizable equilibrium strategy. Finally, Assumption 2 requires which can be satisfied if noting that (this follows from for all ).
8 Conclusion And Future Work
We proposed and studied a strategic model of marketing and product consumption in social networks. Two firms offer substitutable products and compete to maximize the consumption of their products in a social network. Consumers are myopic and update their consumption level as the best response to the consumption of their neighbors in the previous period. This results in linear update dynamics for the product consumption. Moreover, each consumer receives externality from the consumption of each neighbor; the externality is stronger for consumption of the same product. Firms can improve their market share by seeding the key consumers in the market, as their consumption will incentivize the consumption of the same product by their peers given the inter-agent externalities, which in turn can affect the consumption behavior all over the network. We represented the above setting as a duopoly game between the firms and introduced a novel framework that allows for sparse seeding to asymptotically emerge as an equilibrium strategy. We then studied the effect of the network structure on the optimal seeding strategies and the extent to which these strategies could be sparsified, under the proposed equilibrium concept. In particular, we derive necessary and sufficient conditions under which -Nash equilibrium strategies can asymptotically lead to sparse seeding in large populations. The results were demonstrated using a large core-periphery network structure with few highly connected and many sparsely connected nodes. Extending our analysis to time-varying seeding strategies (recurring seeding), networks with uncertainties (e.g., in valuations of consumers about the products), and oligopoly setting are some of the potential venues for future research.
- Amini et al. (2009) Amini, H., M. Draief and M. Lelarge (2009). Marketing in a random network. Network Control and Optimization pp. 17–25.
- Ballester et al. (2006) Ballester, C., A. Calvó-Armengol and Y. Zenou (2006). Who’s who in networks. wanted: the key player. Econometrica 74(5), 1403–1417.
- Bharathi et al. (2007) Bharathi, S., D. Kempe and M. Salek (2007). Competitive influence maximization in social networks. Internet and Network Economics 4858, 306–311.
- Bimpikis et al. (2013) Bimpikis, Kostas, Asuman Ozdaglar and Ercan Yildiz (2013). Competing over networks. submitted for publication.
- Bramoullé et al. (2014) Bramoullé, Yann, Rachel Kranton and Martin D’amours (2014). Strategic interaction and networks. The American Economic Review 104(3), 898–930.
- Candogan et al. (2012) Candogan, Ozan, Kostas Bimpikis and Asuman Ozdaglar (2012). Optimal pricing in networks with externalities. Operations Research 60(4), 883–905.
- Chasparis and Shamma (2010) Chasparis, G.C. and J.S. Shamma (2010). Control of preferences in social networks. In: Proceedings of 49th IEEE Conference on Decision and Control (CDC). pp. 6651–6656.
- Corbo et al. (2007) Corbo, J., A. Calvó-Armengol and D.C. Parkes (2007). The importance of network topology in local contribution games. In: in Proceedings of the 3rd international conference on Internet and network economics. Springer-Verlag. pp. 388–395.
- Daskalakis et al. (2006) Daskalakis, Constantinos, Aranyak Mehta and Christos Papadimitriou (2006). A note on approximate nash equilibria. In: International Workshop on Internet and Network Economics. Springer. pp. 297–306.
- Draief et al. (2006) Draief, M., A. Ganesh and L. Massoulié (2006). Thresholds for virus spread on networks. In: in Proceedings of the 1st international conference on Performance evaluation methodolgies and tools. p. 51.
- Ellison (1993) Ellison, G. (1993). Learning, local interaction, and coordination. Econometrica: Journal of the Econometric Society 61(5), 1047–1071.
- Fazeli et al. (2017) Fazeli, Arastoo, Amir Ajorlou and Ali Jadbabaie (2017). Competitive diffusion in social networks: Quality or seeding?. IEEE Transactions on Control of Network Systems 4(3), 665–675.
- Galeotti and Goyal (2009) Galeotti, A. and S. Goyal (2009). Influencing the influencers: a theory of strategic diffusion. The RAND Journal of Economics 40(3), 509–532.
- Ganesh et al. (2005) Ganesh, A., L. Massoulié and D. Towsley (2005). The effect of network topology on the spread of epidemics. In: INFOCOM 2005. 24th Annual Joint Conference of the IEEE Computer and Communications Societies. Proceedings IEEE. Vol. 2. pp. 1455–1466.
Goyal and Kearns (2012)
Goyal, Sanjeev and Michael Kearns (2012). Competitive contagion in networks.
Proceedings of the 44th symposium on Theory of Computing. pp. 759–774.
- Harsanyi and Selten (1988) Harsanyi, J.C. and R. Selten (1988). A general theory of equilibrium selection in games. MIT Press Books.
- Kandori et al. (1993) Kandori, M., G.J. Mailath and R. Rob (1993). Learning, mutation, and long run equilibria in games. Econometrica: Journal of the Econometric Society 61(1), 29–56.
- Kempe et al. (2003) Kempe, D., J. Kleinberg and É. Tardos (2003). Maximizing the spread of influence through a social network. In: Proceedings of the ninth ACM SIGKDD international conference on Knowledge discovery and data mining. pp. 137–146.
- Kempe et al. (2005) Kempe, D., J. Kleinberg and É. Tardos (2005). Influential nodes in a diffusion model for social networks. Automata, Languages and Programming 3580, 1127–1138.
- Kleinberg (2007) Kleinberg, J. (2007). Cascading behavior in networks: Algorithmic and economic issues. Algorithmic game theory 24, 613–632.
- López-Pintado (2006) López-Pintado, D. (2006). Contagion and coordination in random networks. International Journal of Game Theory 34(3), 371–381.
- Montanari and Saberi (2010) Montanari, A. and A. Saberi (2010). The spread of innovations in social networks. Proceedings of the National Academy of Sciences 107(47), 20196–20201.
- Seeman and Singer (2013) Seeman, Lior and Yaron Singer (2013). Adaptive seeding in social networks. In: Foundations of Computer Science (FOCS), 2013 IEEE 54th Annual Symposium on. IEEE. pp. 459–468.
- Vetta (2002) Vetta, A. (2002). Nash equilibria in competitive societies, with applications to facility location, traffic routing and auctions. In: The 43rd Annual IEEE Symposium on Foundations of Computer Science. pp. 416–425.
- Young (2002) Young, H Peyton (2002). The diffusion of innovations in social networks. Economy as an Evolving Complex System. Proceedings volume in the Santa Fe Institute studies in the sciences of complexity 3, 267–282.
- Young (1993) Young, H.P. (1993). The evolution of conventions. Econometrica: Journal of the Econometric Society 61(1), 57–84.
- Young (2001) Young, H.P. (2001). Individual strategy and social structure: An evolutionary theory of institutions. Princeton University Press.