Junyuan Lin, assistant professor of mathematics at Loyola Marymount University, will investigate “Spectral Algorithms for Dynamic Social Networks and Knowledge Graphs” thanks to $250,000 in grant funding recently awarded to her by the U.S. National Science Foundation (NSF).
Lin’s two-year research project uses her expertise in computational graph theory, machine learning, and data mining. This particular project involves data mining large data sets from social media networks (in this case, X, formerly known as Twitter) to more accurately uncover topic correlations, monitor cluster evolution, and offer insights for future societal discourse, awareness, and policy-making. “We know that misinformation can have devastating consequences. My research aims to track, interpret, and predict patterns of topic evolution and sentiment shifts on online social networks, as well as track, measure, and correlate current and significant global events or movements,” said Lin.
Her previous collaborative and award-winning research involving protein interactive networks in the medical field produced a Diffusion State Distance (DSD) metric proven effective on weighted and directed networks, earning first place in the 2016 DREAM Disease Module Identification Challenge. This research involved developing methods and analysis to classify and cluster proteins in different species to provide a basis for future research into disease etiologies and treatment.
For Lin’s current research on dynamic social networks, there is a scarcity of methods for considering directed and weighted edges for graphing data structure, which are important for addressing the heterogeneous nature of social networks. “Drawing similarities between online social networks and protein interaction networks, we will customize spectral graph metrics based on DSD for heterogeneous social interaction networks and their multi-layer knowledge graphs,” Lin explains.
In the last three years, Lin’s collaborative research team has processed a dataset with approximately 700 million tweets from the internet and was able to define what trends in topics are being highly discussed online. As a result, these preliminary findings have led to the publication of several papers in IEEE Big Data on the topics of COVID-19, QAnon, and hate speech. “My hope is to use mathematical models and algorithms to further investigate these trends on heated topics and predict possible new topics born out of trends, such as in the magnitude of the COVID-19 pandemic or global hate crimes,” Lin explains. The goal is to document the sentiments surrounding these topics and hopefully the data is useful for policy makers to understand people’s intentions and possible future actions.”
Lin currently has four LMU students collaborating on this research including three undergraduate applied mathematics or statistics majors and one second-year graduate student majoring in computer science. “Going forward, I also plan to collaborate with faculty and students from other departments, including computer science, mathematics, and sociology, to ensure user-friendly access to this knowledge graph framework and spectral algorithms,” she said. “I also hope to create interdisciplinary courses to expose students to research areas in data mining on social networks.”
While Lin believes research and analysis of discourse within dynamic social networks is a long-term proposition, she envisions many applications and is considering studying other types of social networks with human interaction such as podcasts and platforms such as Reddit. She is also developing different algorithms to efficiently solve questions when working with huge data networks. She wants to ensure the technology she is using is transparent and can be an open-source tool for others to use in a variety of research applications.
“I want to provide more opportunities for LMU students to be involved in my investigative work which overall focuses on building efficient models for huge data sets. We then analyze this big data, learn about patterns, and build algorithms with predictive models using machine learning,” said Lin. “While my research can be applied across many fields, it is connected by the use of scientific computing and spectral analysis on complex, large networks.” The grant was awarded in August 2024 by NSF’s Division of Mathematical Sciences. Research proposals evaluated by the NSF use the organization’s two board-approved merit review criteria: intellectual merit and broader impacts, which recognize both the research and the project’s potential impact on society. NSF proposal reviewers may also look at other factors such as different approaches to significant research and education questions, potential for transformational advances in a field, and capacity for building in a new or promising research area.
