SOCAS: A Social Media Public Opinion Causal Analysis System Validated by the Funnel of Causality Theory

Authors

Chen Yi (Fudan University), Yihang Yang (Fudan University), Xingyu Lan (Fudan University), Siming Chen (Fudan University)

Presentation

Session: My followers need to know about this!
Time: Thursday, Nov 12, 10:36 – 10:48 (US/Eastern) · session 10:00 – 11:30
Location: Hall America center

Keywords

Visual analytics, causal relationship extraction, social media analysis, causal funnel theory, zero-shot learning, structural sensitivity analysis

Abstract

Social media has become a central platform for public discourse, yet understanding the causal mechanisms underlying opinion dynamics remains challenging due to semantic noise, causal hallucinations, and the lack of theoretical grounding in existing computational approaches. We present SOCAS, a visual analytics framework that bridges unstructured social media discourse and interpretable causal relationship extraction(CRE). Grounded in Causal Funnel Theory, SOCAS introduces a hierarchical transition model that organizes influencing factors from distal exogenous conditions to proximal political actions. To support scalable analysis, we develop a zero-shot CRE pipeline leveraging Large Language Models (LLMs), enabling domain-agnostic extraction of event units and causal relationships, with confidence scores and temporal post-filtering to mitigate hallucinations. To enhance interpretability and user engagement, SOCAS incorporates an interactive structural sensitivity analysis module that supports structural perturbation reasoning and dynamic exploration of causal structures, complemented by a natural language interface(NLI) for conversational refinement. In addition, we design specialized analytical modules to trace complex behavioral mechanisms, such as communication and political translation.

For Practitioners

Target practitioners include data journalists, policy and intelligence analysts, social media researchers, and data scientists building LLM-based extraction pipelines.