HINTS: Uncovering Human Factors in Time Series Forecasting

This paper introduces HINTS, a self-supervised learning framework that extracts human factors from time series data for improved forecasting. The key innovation is the ability to do this without relying on external data sources, which reduces data dependency costs. The use of the Friedkin-Johnsen (FJ) opinion dynamics model as a structural inductive bias is a novel approach. The paper's strength lies in its potential to improve forecasting accuracy and provide interpretable insights into the underlying human factors driving market dynamics.

• •Proposes HINTS, a self-supervised framework for extracting human factors from time series data.
• •Avoids reliance on external data sources, reducing data dependency costs.
• •Employs the Friedkin-Johnsen (FJ) opinion dynamics model as a structural inductive bias.
• •Demonstrates improved forecasting accuracy and interpretability through experiments and case studies.