Enhanced Social Event Detection through Dynamically Weighted Meta-Paths Modeling

Social event detection (SED) identifies significant events in social networks by analyzing complex interactions using both textual data and structural relationships that often span multiple nodes. This requires exploring long-range dependencies, which increases computational costs, especially with many neighbors.Therefore, in this paper, we present the Dynamically Weighted Meta-Paths Modeling (DWMM) framework for large-scale social event detection. It includes three main modules: 1) a graph building module to convert social event data into Heterogeneous Information Networks (HINs); 2) a meta-path searching module to determine the significant meta-paths and their importance; 3) a model training module that uses weighted top-k meta-paths for social event detection. Extensive experiments on three widely used social event detection datasets show that DWMM enhances performance, demonstrating its effectiveness.