Sobre o comportamento de agentes racionais em redes complexas

Abstract

The constant advancement of information systems allows, at a growing rate, more data to be generated and stored. From routine situations, such as phone conversations, to the actu-ation of highly complex technological systems, such as Wireless Sensor Networks (WSNs) to detect weather events, data are generated and stored registering every action and decision made by the agents of these systems. It is fascinating that, behind these records, we see the reflection of the environment itself, since behind every record, there is a decision made by some entity. Therefore, the knowledge of how to process this valuable and very large evolv-ing database can lead, consequently, to a better understanding of the interests and dynamics of each entity in a determined environment or in the society. In this thesis, we focus on systems that are made up of entities (individuals, organi-zations and computational systems) capable of interacting among themselves in a rational way, reflecting their interests and activity dynamics. We call these systems Decision-basedComplex Networks (DBCN), these individuals and/or organizations, nodes or agents, and the interactions between nodes, edges. The main characteristic of a DBCN is that it evolves according to the personal motivations of its nodes. Therefore, the main objective of this the-sis is to analyze the behavior of agents of DBCN. We plan to observe real and hypothetical scenarios where decisions play an important role in the evolution of the network. When we fully understand the motivations behind the actions of the agents, we may be able to model, predict and control their behavior. We divide the main objetive of this thesis in three spe-cific goals: (i) Modeling, where we aim to accurately represent the behavior of the agents in a DBCN; (ii) Predicting, where we aim to predict how the system will evolve, and (iii) Controlling, where we aim to be able to make the agents to act according to a determined goal. First, we show a compact analysis of the three aspects of DBCNs we tackle in thisthesis, i.e., modeling, predicting and controlling, in three real-world datasets describing user mobility activity. Then, we focus solely on modeling communication networks. We pro-posed models for the individual behavior of the users of this network and, from this, weproposed applications for data summarization, anomaly detection and network monitoring. Then, we focus on predicting in competitive networks formed from sports leagues such as the North American National Basketball Association (NBA) and the Major League Baseball (MLB). We proposed a prediction model that can be used to identify likely teams to win and to fail in a following season. Finally, we focus on controlling, investigating decision making in computer networks, in a scenario where a group of WSNs are deployed in the same re-gion and they may interact with themselves to request or share computational resources. Wemodeled the problem of cooperation among two or more different WSNs by the concepts of game theory and from it we present, as a control application, a protocol to allow cooperation among them.