Self Adaptation of Cooperation in Multi-agent Content Sharing Systems

This paper considers an adaptive data dissemination scenario that applies an autonomic trust protocol to a network of agents. The protocol uses social network structures to incentivize cooperation. Validation is conducted through simulation of content sharing between peers which uses similarity of interest between peers to define payoff. Positive correlation is observed between the number of social links placed and payoff received by single agents. Content sharing allows calculation of similarity between agents within a system. Prior interaction history drives the formation of social links between nodes and allows estimation of an individuals cooperation by another. Agents may adaptively change their cooperation levels when forming social relation-ships by copying those of the most `popular´ members of their own social groups. Adaptation mechanisms can be prioritized within communities sharing similar interests. Similarity of interest communities and their initial cooperation levels both have an effect on the self-adaptation of cooperation. The most divergent and least cooperative nodes have fewer opportunities to form new social links, increase their cooperation levels, and consequently increase their payoff. Self-adaptation results in higher payoff for the population compared to the static scenario in which adaptation of agents cooperation does not occur.