Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SCTc-TE: A Comprehensive Formulation and Benchmark for Temporal Event Forecasting

Yunshan Ma, Chenchen Ye, Zijian Wu, Xiang Wang, Yixin Cao, Liang Pang, Tat-Seng Chua

TL;DR

This work introduces SCTc-TE, a structured, complex, and time-complete formulation for temporal event forecasting, and presents a fully automated pipeline to construct large SCTc-TE datasets (MidEast-TE and GDELT-TE) from news. It then proposes LoGo, a forecasting model that fuses local context within a complex event and global context across all events, using RT-Mod for relational-temporal representation and a ConvTransE decoder. Empirical results on both datasets show LoGo substantially outperforms static KG and temporal KG baselines, with ablations highlighting the importance of dual-context fusion and early integration. The authors release code and datasets, enabling broader exploration and practical impact in structured temporal event forecasting.

Abstract

Temporal complex event forecasting aims to predict the future events given the observed events from history. Most formulations of temporal complex event are unstructured or without extensive temporal information, resulting in inferior representations and limited forecasting capabilities. To bridge these gaps, we innovatively introduce the formulation of Structured, Complex, and Time-complete temporal event (SCTc-TE). Following this comprehensive formulation, we develop a fully automated pipeline and construct a large-scale dataset named MidEast-TE from about 0.6 million news articles. This dataset focuses on the cooperation and conflict events among countries mainly in the MidEast region from 2015 to 2022. Not limited to the dataset construction, more importantly, we advance the forecasting methods by discriminating the crucial roles of various contextual information, i.e., local and global contexts. Thereby, we propose a novel method LoGo that is able to take advantage of both Local and Global contexts for SCTc-TE forecasting. We evaluate our proposed approach on both our proposed MidEast-TE dataset and the original GDELT-TE dataset. Experimental results demonstrate the effectiveness of our forecasting model LoGo. The code and datasets are released via https://github.com/yecchen/GDELT-ComplexEvent.

SCTc-TE: A Comprehensive Formulation and Benchmark for Temporal Event Forecasting

TL;DR

This work introduces SCTc-TE, a structured, complex, and time-complete formulation for temporal event forecasting, and presents a fully automated pipeline to construct large SCTc-TE datasets (MidEast-TE and GDELT-TE) from news. It then proposes LoGo, a forecasting model that fuses local context within a complex event and global context across all events, using RT-Mod for relational-temporal representation and a ConvTransE decoder. Empirical results on both datasets show LoGo substantially outperforms static KG and temporal KG baselines, with ablations highlighting the importance of dual-context fusion and early integration. The authors release code and datasets, enabling broader exploration and practical impact in structured temporal event forecasting.

Abstract

Temporal complex event forecasting aims to predict the future events given the observed events from history. Most formulations of temporal complex event are unstructured or without extensive temporal information, resulting in inferior representations and limited forecasting capabilities. To bridge these gaps, we innovatively introduce the formulation of Structured, Complex, and Time-complete temporal event (SCTc-TE). Following this comprehensive formulation, we develop a fully automated pipeline and construct a large-scale dataset named MidEast-TE from about 0.6 million news articles. This dataset focuses on the cooperation and conflict events among countries mainly in the MidEast region from 2015 to 2022. Not limited to the dataset construction, more importantly, we advance the forecasting methods by discriminating the crucial roles of various contextual information, i.e., local and global contexts. Thereby, we propose a novel method LoGo that is able to take advantage of both Local and Global contexts for SCTc-TE forecasting. We evaluate our proposed approach on both our proposed MidEast-TE dataset and the original GDELT-TE dataset. Experimental results demonstrate the effectiveness of our forecasting model LoGo. The code and datasets are released via https://github.com/yecchen/GDELT-ComplexEvent.
Paper Structure (23 sections, 6 equations, 6 figures, 7 tables)

This paper contains 23 sections, 6 equations, 6 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Problem Formulation and Dataset
  3. Problem Formulation
  4. SCTc-TE Construction Pipeline
  5. Domain and Data Sources
  6. Structured Event Extraction
  7. Complex Event Identification
  8. Dataset Evaluation
  9. Statistics
  10. Evaluation
  11. Methodology
  12. SCT Event Forecasting
  13. Local and Global Context Modeling
  14. Prediction and Optimization
  15. Experiments
  16. ...and 8 more sections

Figures (6)

  • Figure 1: (a) Comparison between our SCTc-TE and previous TE formulations; (b) An illustration of our SCTc-TE formulation; (c) The motivation of leveraging both local and global contexts.
  • Figure 2: The dataset construction pipeline of SCTs-TE consists of: 1) Complex Event Identification clusters complex events by time-aware document embedding and 2) Structured Event Extraction hierarchically extracts structured events with LLMs.
  • Figure 3: Illustration of ten examples of CEs, each with its CE ID, wordcloud, timespan, number of involved news articles, entities, and atomic events.
  • Figure 4: The overall framework of this work consists of two parts: 1) the SCTc-TE construction pipeline (the top left part), and 2) the SCTc-TE forecasting model LoGo (the right and bottom part).
  • Figure 5: Sub-figure (a,b,c) show the effects of context historical length, and (d,e) study the effects of GNN layers.
  • ...and 1 more figures