Potential Indirect Relationships in Productive Networks

Abstract

Productive Networks, such as Social Networks Services, organize evidence about human behavior. This evidence is independent of the network content type, and may support the discovery of new relationships between users and content, or with other users. These indirect relationships are important for recommendation systems, and systems where potential relationships between users and content (e.g., locations) is relevant, such as with the emergency management domain, where the discovery of relationships between users and locations on productive networks may enable the identification of population density variations, increasing the accuracy of emergency alerts. This thesis presents a Productive Networks model, which enables the development of a methodology for indirect relationships discovery, using the metadata on the network, and avoiding the computational cost of content analysis. We designed and conducted a set of experiments to evaluate our proposals. Our results are twofold: firstly, the productive network model is sufficiently robust to represent a wide range of networks; secondly, the indirect relationship discovery methodology successfully identifies relevant relationships between users and content. We also present applications of the model and methodology in several contexts.