EventConnector: Mining Social Event Relations through Temporal Graphs

ACL ARR 2025 May Submission7318 Authors

Abstract: Understanding and retrieving related real-world events based on their temporal dynamics is a fundamental challenge in time-sensitive applications such as forecasting, information retrieval, and social analysis. Existing methods often rely on semantic similarity or global time-series alignment, which overlook the transient and directional dependencies that frequently underlie real-world correlations. In this work, we introduce \textbf{EventConnector}, a general framework for constructing a temporal event graph that captures localized co-fluctuations and lead-lag relationships between events through their time-series trajectories. The resulting graph encodes both synchronous activity and directional influence, enabling the discovery of non-obvious, cross-domain associations. To further enrich the graph structure, we incorporate a multi-hop detection mechanism that reveals transitive temporal dependencies. Experiments on real-world prediction market data show that EventConnector uncovers non-trivial temporal structures and achieves a substantial 18.89\% improvement in event retrieval and time-series forecasting tasks under limited supervision. These results highlight the effectiveness of temporal graph modeling in capturing latent event relationships beyond what semantic similarity or traditional alignment techniques can offer.

Paper Type: Long

Research Area: Information Extraction

Research Area Keywords: time series forecasting, temporal graph, social modeling, deep neural network, data mining, prediction task

Contribution Types: Data analysis

Languages Studied: English

Submission Number: 7318