Partners and International Organizations
(English)
|
AT, BE, BT, CH, CZ, DE, DK, ES, FI, FR, GR, HU, IE, IT, LT, MK, NL, NO, PL, RO, RS, TR, UK
|
Abstract
(English)
|
How do agents become influential in social and economic networks? Certainly not only because they were 'born' in a favorable social or economic environment, but mainly because of their private effort to improve their given position over time. In this project, we try to understand agent's strategies to improve their position and potential conflicts resulting from these strategies both empirically and by means of theoretical models. Agent's influence is measured by their centrality, which is an established measure in complex network theory to reflect how important an agent is in connecting other agents. If an agent is able to get connected to agents with a high centrality, this increases its centrality as well. Therefore, agents directly or indirectly compete to establish links to more central agents. If the number of links is constrained by some capacity restrictions, this also implies that links to less central agents are removed. Thus, from the perspective of the agent it is a strategic decision to whom to connect, in order to increase its own centrality. In this project we focus on (i) the topological properties of the global network resulting from such strategies, and (ii) on the emergence of hierarchies in the agent organization in the course of time. We further try to obtain empirical measures of agent's centrality from social and economic data, to infer their strategies. From the theoretical investigations, we concluded that emerging networks have a strong hierarchical structure described as a 'nested-split graph'. This is characterized by a short path length, a high clustering coefficient, a power-law tail in the degree distribution, and dissortativity, i.e. agents with high degree are mostly connected to agents with low degree. In the presence of capacity constraints, however, networks loose their 'nested-split graph' topology. Their degree distribution changes into an exponential distribution and they can become assortative, i.e. agents with high degree are mostly connected to other agents with high degree. On the empirical side, we investigated large datasets from social networking sites (Dooyoo, Epinions), in which users can post reviews about products, rate reviews written by others and form social links with others. The centrality of users was obtained by the $k$-shell decomposition method, which assigns each user to a ``shell'' that indicates its importance. We then tried to identify the profile of the more central users, to understand whether there is a common pattern in their behavior, or strategy. We identified that users became more central by writing mostly critical product reviews or by writing reviews that others regarded as most helpful. We also found that centrality has a positive feedback on the user's activity, i.e. the time spent in the social network and the length of the reports written. This may further increase the user's centrality as other users are inclined to establish a social link to her. In Dooyoo, there are monetary incentives for users to compete actively for increasing their centrality by getting friendship links and acknowledgement of their reviews, in Epinions there is at least an indirect competition for attention.
|
