|dc.description.abstract||The advancement of complex social systems such as Facebook and Twitter has led to huge volume of user generated contents, which enable detailed tracking and characterization of human activities. Users of these systems interact with each other to make decisions in events such as information dissemination and to learn knowledge such as rating of online businesses. Quantitative analysis and comprehension of mechanisms of users' behaviors in these systems are both intriguing and imperative in many academic fields (e.g., economics and social/political sciences) and applications (e.g., online advertisement and management of electronic commerce). In addition, due to the high commercial and research values of these user generated data for many individuals and companies, various data trading platforms are emerging to facilitate data transactions and to extract remarkable profits from the data markets, yet few methodological trading schemes are available in the literature.
Therefore, in this dissertation, we are motivated to examine users' behaviors during several learning and decision making processes in networked systems and to design an efficient data trading mechanism systematically for markets with multiple data agents. Specifically, we first propose a graphical evolutionary game theoretic framework for information propagation over heterogeneous networks and analytically study the dynamics and stable states of the game. Theoretical results are corroborated by numerical experiments on real-world information diffusion data. Secondly, to incorporate users' long-term incentives, we propose a sequential game to model the decision-making procedures in generic popularity dynamics. Properties of the symmetric Nash equilibrium of the game are theoretically analyzed and match well with empirical observations from real world popularity dynamics such as information diffusion dynamics and paper citation dynamics. Thirdly, an evolutionary game theoretic learning algorithm is proposed for the social learning problem, where networked agents collaborate to detect some unknown system state. Theoretical analysis manifests that the stable states of the proposed distributed learning algorithm coincide with the decisions of a fictitious centralized detector. Lastly, we investigate the data trading problem in a market with multiple data owners, collectors and users. An efficient data trading mechanism based on iterative auctions is presented and we demonstrate that the mechanism converges to the socially optimal operation point and possesses appealing economic properties. Numerical studies based on data prices of real-world data transaction platforms are shown to verify the effectiveness of the proposed trading mechanisms.||en_US