Data Science Talk Series 

How do LLMs empower Computational Social Science?  (Online)

Abstract: Computational Social Science (CSS) increasingly relies on large-scale, multimodal, and high-dimensional data to understand complex social phenomena. However, CSS faces persistent challenges, including unstructured and noisy data, annotation bottlenecks, multimodal integration, and feature-rich inference. This talk presents a research program on how large language models (LLMs) can empower CSS tasks. It demonstrates how LLM agents address core CSS challenges through real-world case studies: extracting structured knowledge from noisy social media for drug side-effect surveillance, augmenting or automating content annotation for hate and toxicity detection, simulating human-perceived experiences during extreme events through multimodal data fusion of geospatial and socioeconomic signals, and enabling complex decision-making via multi-agent reasoning in business partner selection. Collectively, these examples illustrate how LLMs can transform traditional CSS pipelines from static analysis tools into adaptive, reasoning-driven systems capable of scalable inference and richer modeling of social processes. 

Here is the paper list associated with the talk:

• • Crowdsourcing-Based Knowledge Graph Construction for Drug Side Effects Using Large Language Models with an Application on Semaglutide; 2025 AMIA. https://pubmed.ncbi.nlm.nih.gov/41726410/ 

• • “HOT” ChatGPT: The Promise of ChatGPT in Detecting and Discriminating Hateful, Offensive, and Toxic Comments on Social Media. 2024 ACM TWEB. https://doi.org/10.1145/3643829 

• • LLMs as World Models: Data-Driven and Human-Centered Pre-Event Simulation for Disaster Impact Assessment. 2026 EMNLP.  https://aclanthology.org/2025.emnlp-main.153/.

• • PartnerMAS: An LLM Hierarchical Multi-Agent Framework for Business Partner Selection on High-Dimensional Features. 2025 ArXiv. https://doi.org/10.48550/arXiv.2509.24046

Bio: Dr. Lingyao Li is an Assistant Professor in the School of Information at the University of South Florida. He completed his postdoctoral training at the University of Michigan School of Information and received his Ph.D. in Civil and Environmental Engineering from the University of Maryland, College Park. His research integrates artificial intelligence (AI), particularly large language models (LLMs), with crowdsourced data from social media and mobile devices to address socio-technical challenges in computational social science, with a focus on urban environments and healthcare. A key topic of his work is the concept of citizens-as-sensors, using large-scale, real-world data to study community resilience, disaster impacts, public emotional and behavioral responses, and urban accessibility. More recently, his research has focused on the capabilities and design of LLM-based multi-agent systems in these domains, as well as the trustworthiness of LLMs for information retrieval and decision support. His work has been published in leading venues including AAAI, ACL, EMNLP, WWW, CHI, ICWSM, AMIA, ACM TWEB, ACM TIST, ASCE journals, International Journal of Information Management, Computers & Education, Sustainable Cities and Society, Computers Environment & Urban Systems, International Journal of Disaster Risk Reduction, Journal of Biomedical Informatics, and Journal of Medical Internet Research.