TempReasoner: neural temporal graph networks for event timeline construction.

Area of Science:

• Artificial Intelligence
• Knowledge Extraction
• Temporal Reasoning

Background:

• Automated event timeline construction from unstructured temporal data is challenging.
• Existing methods struggle with fine-grained temporal dependencies, sparse event interactions, and causal ordering across domains.

Purpose of the Study:

• To propose TempReasoner, a novel neural temporal graph network for building automated event timelines.
• To address limitations in existing temporal reasoning methods for complex datasets.

Main Methods:

• Utilizes a neural temporal graph network with dynamic spatio-temporal attention and reinforced temporal reasoning.
• Combines temporal knowledge graphs with adaptive graph neural networks and a multi-scale temporal attention model.
• Employs a hierarchical temporal encoder with gated recurrent units and a novel temporal consistency loss.

Main Results:

• Achieves 94.3% accuracy in ordering event timelines across five benchmark datasets.
• Operates in real-time with an average latency of 127 ms per event sequence.
• Demonstrates robust performance across legal, news, and biomedical event analysis.

Conclusions:

• TempReasoner effectively constructs automated event timelines by jointly modeling local and global temporal patterns.
• The system offers a scalable and efficient solution for temporal reasoning in various enterprise applications.
• The proposed architecture enhances temporal coherence and accuracy in knowledge extraction tasks.