What Can Large Language Models Tell Us about Time Series Analysis

Link:https://arxiv.org/pdf/2402.02713.pdf#page=9&zoom=100,384,533

Abstract

LLM can get involved in time series analysis.

LLM Background

- Pro: 0 shot 
- Con: Primary for Text sequence

Idea: time series and textual information provide essential contexts, LLMs contribute internal knowledge and reasoning capabilities, and pre-trained time series models offer fundamental pattern recognition assurances

Time series data modality:

• Univariate: X = {x1, x2, · · · , xT } ∈ R^T

• Multivariate: X = {x1, x2, · · · , xT } ∈ R^(N*T)