Knowledge acquisition and reconstruction in complex networks

• Ano de defesa: 2023
• Autor(a) principal: Guerreiro, Lucas
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Caminhadas aleatórias; Ciência das redes; Descoberta do conhecimento; Knowledge acquisition; Network science; Network topology; Random walks; Sequences; Sequências; Topologia de redes
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-31082023-084426/

Complex networks have been used in several applications in the past few decades. Complex systems can be observed in applications such as transportation, energy systems, internet, biology, and logistics, among other possible *implementations*. In such structures, there may be agents exploring nodes and identifying new concepts and discovering the network; this kind of exploration is known as Knowledge Acquisition and it has been deeply researched for the past decades. When exploring a network, i.e., acquiring knowledge in it, the path explored by an agent can be seen as a sequence of visited nodes. In this context, this thesis allowed us to observe the behavior of different network dynamics and topologies when acquiring knowledge. Moreover, using machine learning techniques, we have proposed a framework that showed it is possible to recover the generating structures that constructed an unknown sequence. Finally, we have evaluated how global properties of a network are reflected in structures generated by sequences. Thus, we have presented an analysis whether local information of a network are biased or it is indeed a real picture of the network as a whole; this analysis allowed us to measure the impact of the sequences size while identifying networks properties. Hence, the results presented in this thesis have shown the behavior of different structures during the knowledge acquisition process. Lastly, we can highlight the framework built in this work, which allowed to classify which are the original topology and dynamics that generated a given sequence. Such results may enable several applications in network science, and pave the knowledge in this area. Among the main results, this work has allowed the proper identification of sequences generating structures from the properties obtained during the reconstruction of such sequences as a complex network; and, moreover, it was possible to observe that small sequences allow the identification of the structures with high accuracyComplex networks have been used in several applications in the past few decades. Complex systems can be observed in applications such as transportation, energy systems, internet, biology, and logistics, among other possible *implementations*. In such structures, there may be agents exploring nodes and identifying new concepts and discovering the network; this kind of exploration is known as Knowledge Acquisition and it has been deeply researched for the past decades. When exploring a network, i.e., acquiring knowledge in it, the path explored by an agent can be seen as a sequence of visited nodes. In this context, this thesis allowed us to observe the behavior of different network dynamics and topologies when acquiring knowledge. Moreover, using machine learning techniques, we have proposed a framework that showed it is possible to recover the generating structures that constructed an unknown sequence. Finally, we have evaluated how global properties of a network are reflected in structures generated by sequences. Thus, we have presented an analysis whether local information of a network are biased or it is indeed a real picture of the network as a whole; this analysis allowed us to measure the impact of the sequences size while identifying networks properties. Hence, the results presented in this thesis have shown the behavior of different structures during the knowledge acquisition process. Lastly, we can highlight the framework built in this work, which allowed to classify which are the original topology and dynamics that generated a given sequence. Such results may enable several applications in network science, and pave the knowledge in this area. Among the main results, this work has allowed the proper identification of sequences generating structures from the properties obtained during the reconstruction of such sequences as a complex network; and, moreover, it was possible to observe that small sequences allow the identification of the structures with high accuracy