Individual-centered Partial Information in the Latent Space Network Models

Description

Network data is critical in quantitative and qualitative analyses across diverse fields such as social science, biology, and civil engineering. It illuminates complex relationships between subjects, offering an understanding beyond the individual attributes captured by traditional datasets. Specifically, in social network studies, nodes and edges represent social actors and their connections. As technological advancements expand networks rapidly, managing and extracting insights from these networks present challenges for statisticians.Statistical network research often estimates graph features using complete network data or information aggregated from multiple subgraphs, while individuals often possess limited knowledge extending beyond their immediate connections. Addressing this discrepancy, recent research has introduced an individual-centered partial information framework to characterize individual local views of the entire network and enable the development of methods using an individual-centered perspective to study global structures of social networks. The existing work under this framework primarily concentrated on inferring community memberships and relied on restrictive structural assumptions.Our project focuses on analyzing individual-centered partial information within general network model settings. The primary aim is to explore an individual’s potential to study the global properties of a general social network. We will use latent space models’ flexibility and visualization compatibility to develop methods that offer a visually appealing and interpretable model-based representation of network relationships with partial information. The developed methods will not only demonstrate robust performance in tasks including community detection and social distance measurement, but also possess advantageous theoretical properties without necessitating stringent structural assumptions. The project’s results can contribute to various network studies, e.g., social media influencer/live streamer market analysis, visualization of statistics community’s influence, and traffic volume prediction.Traditionally, the substantial costs and complexities associated with network data collection, acquisition, and computation of large network properties have meant that only large institutions and research groups could afford to conduct network analysis to understand their target societies. Our project will provide efficient methods enabling individuals to derive insights from global network properties using their personal data on their devices. This approach significantly reduces the costs related to data acquisition and large-matrix computation, while concurrently ensuring personal data privacy. By doing so, we hope to extend the benefits of network analysis to a broader audience, including individuals, small businesses, and independent researchers, thereby promoting a more inclusive, equitable, and secure information ecosystem.