BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//UNIFR/WEBMASTER//NONSGML v1.0//EN
CALSCALE:GREGORIAN
BEGIN:VEVENT
DTSTART;VALUE=DATE:20231023T160000
DTEND;VALUE=DATE:20231023T160000
UID:14578@agenda.unifr.ch
DESCRIPTION:Complex networks provide a flexible way to represent \nheterogeneous connectivity and interactions within real-world \nsystems. Nowadays, people's lives are immersed in a world full of \ndiverse and intricate complex networks. These networks not only \nprovide conveniences that benefit our daily lives, such as enabling \nsocial networking, personalized recommendations, and shopping \npromotions. But they also bring certain negative impacts, such as \nfacilitating the spread of misinformation, rumors, and infectious \ndiseases. Therefore, developing a comprehensive understanding \nof the formation mechanisms, structures, and dynamics of \ncomplex networks in these socio-technical systems remains a \npressing and challenging scientific endeavor. This thesis aims to \nexplore multiple aspects of the applications of complex network \nanalysis, with the following main contributions:\nFirst, network dismantling. Network dismantling strategies will be \ncritical for managing the spread of COVID-19. Public health \ninterventions like social distancing and creating social bubbles aim \nto reduce infections by limiting interactions between social \nnetworks, thus reducing exposure risk. These measures essentially \nreconfigure the connections in the underlying contact networks. \nFrom a theoretical perspective, determining the optimal strategy \nto fragment networks to disrupt disease transmission maps to the \noptimal bond percolation (OBP) problem on networks. OBP \ninvolves selectively removing network connections to fragment the \nnetwork into small components to minimize the size of the giant \nconnected component. This has direct relevance for targeted \nsocial distancing policies that try to maximally disrupt the contact \nnetwork with minimal disruption to individuals. In Chapter 4, I will \nformally introduce the OBP problem and discuss several heuristic \nstrategies and benchmark algorithms.\nSecond, a reputation system based on a network/bipartite graph.\nAs online platforms like e-commerce, review sites, and social \nmedia grow, effectively evaluating quality and reputation through \nrating systems has become critical across many domains. \nTypically, an item's reputation is quantified by aggregating ratings \nand reviews left by users based on their experiences. However, \nuser-contributed ratings can be intentionally misleading or biased, \nfailing to reflect true underlying quality. In particular, certain users \nmay exhibit consistent biases in their ratings across items, like being \noverly critical or lenient. Other users may be motivated to boost or \nattack specific targets. Such rating biases can significantly distort \nreputation systems and lead to poor recommendations if left \nunaccounted for. To address this challenge, evaluating the \nreputation of raters themselves becomes vital. We need scalable \ncomputational methods to infer which users provide accurate and \nreliable ratings versus those who skew ratings up or down. By \ncorrecting for rating biases, we can better reveal the intrinsic \nquality of items. In Chapter 5, we introduce an iterative balancing \n(IB) model that alternates between estimating user reputation and \ninferring unbiased item quality from the collective ratings. \nExperiments on movie rating data demonstrate the IB approach \nyields highly consistent results across iterations and effectively filters \nout different types of synthetic rating biases.\nThird, information popularity prediction on social networks. The \nability to predict the size of information cascades in online social \nnetworks is critical for a variety of applications, including decisionmaking and viral marketing. Traditional methods either rely on \ncomplex time-varying features, which are difficult to extract from \nmultilingual and cross-platform content, or rely on network \nstructures and properties, which are often difficult to obtain. To \naddress these issues, we conducted an empirical study using data \nfrom two well-known social networking platforms (WeChat and \nWeibo). The findings suggest that the information cascade process \nis best described as an activation-decay dynamic. Based on this, \nwe develop an Activate-Decay (AD)-based algorithm that can \naccurately predict the long-term popularity of online content \nbased only on the number of early retweets. An indirect \nrelationship between the peak retweeting amount and the total \npropagation amount was also found. Finding the peak of \ninformation diffusion can significantly improve the prediction \naccuracy of the model. This content will be introduced in detail in \nChapter 6.\nThe research presented in this thesis provides valuable insights into \nthe function and influence of complex networks across various \nreal-world systems.
SUMMARY:Complex Networks and Their Impact: Analysis and Applications
CATEGORIES:Soutenance de mémoire/thèse
LOCATION:PER 08\, 2.73\, Chemin du Musée 3\, 1700 Fribourg
URL;VALUE=URI:https://agenda.unifr.ch/e/fr/14578
END:VEVENT
END:VCALENDAR