NUM2EVENT: Interpretable Event Reasoning from Numerical time-series

TL;DR

This work tackles interpreting numerical time-series when contemporaneous text is unavailable by formulating number-to-event decoding and proposing a framework that maps numbers to structured AAOD events. The approach combines an Agent-Guided Event Extractor (AGE) for semantically coherent supervision, an Event-Driven Time-Series Generator (EveDTS) based on a marked multivariate Hawkes process to synthesize aligned data, and a two-stage time-series encoder–LLM fine-tuning pipeline that yields interpretable reasoning traces linking numeric changes to events, outputting simultaneous event hypotheses. Evaluations on Energy and Public Health datasets show substantial gains in event-level precision and recall over strong LLM baselines, with ablations confirming the value of synthetic data, the numeric encoder, and staged fine-tuning. The method demonstrates a path to bridge quantitative reasoning and semantic understanding, enabling explainable, text-free event decoding directly from numerical dynamics in timely decision-making contexts.

Abstract

Large language models (LLMs) have recently demonstrated impressive multimodal reasoning capabilities, yet their understanding of purely numerical time-series signals remains limited. Existing approaches mainly focus on forecasting or trend description, without uncovering the latent events that drive numerical changes or explaining the reasoning process behind them. In this work, we introduce the task of number-to-event reasoning and decoding, which aims to infer interpretable structured events from numerical inputs, even when current text is unavailable. To address the data scarcity and semantic alignment challenges, we propose a reasoning-aware framework that integrates an agent-guided event extractor (AGE), a marked multivariate Hawkes-based synthetic generator (EveDTS), and a two-stage fine-tuning pipeline combining a time-series encoder with a structured decoder. Our model explicitly reasons over numerical changes, generates intermediate explanations, and outputs structured event hypotheses. Experiments on multi-domain datasets show that our method substantially outperforms strong LLM baselines in event-level precision and recall. These results suggest a new direction for bridging quantitative reasoning and semantic understanding, enabling LLMs to explain and predict events directly from numerical dynamics.