Saved in:
Bibliographic Details
Main Authors: Kim, Kai, Tsai, Howard, Sen, Rajat, Das, Abhimanyu, Zhou, Zihao, Tanpure, Abhishek, Luo, Mathew, Yu, Rose
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2411.06735
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929599468797952
author Kim, Kai
Tsai, Howard
Sen, Rajat
Das, Abhimanyu
Zhou, Zihao
Tanpure, Abhishek
Luo, Mathew
Yu, Rose
author_facet Kim, Kai
Tsai, Howard
Sen, Rajat
Das, Abhimanyu
Zhou, Zihao
Tanpure, Abhishek
Luo, Mathew
Yu, Rose
contents Current forecasting approaches are largely unimodal and ignore the rich textual data that often accompany the time series due to lack of well-curated multimodal benchmark dataset. In this work, we develop TimeText Corpus (TTC), a carefully curated, time-aligned text and time dataset for multimodal forecasting. Our dataset is composed of sequences of numbers and text aligned to timestamps, and includes data from two different domains: climate science and healthcare. Our data is a significant contribution to the rare selection of available multimodal datasets. We also propose the Hybrid Multi-Modal Forecaster (Hybrid-MMF), a multimodal LLM that jointly forecasts both text and time series data using shared embeddings. However, contrary to our expectations, our Hybrid-MMF model does not outperform existing baselines in our experiments. This negative result highlights the challenges inherent in multimodal forecasting. Our code and data are available at https://github.com/Rose-STL-Lab/Multimodal_ Forecasting.
format Preprint
id arxiv_https___arxiv_org_abs_2411_06735
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Multi-Modal Forecaster: Jointly Predicting Time Series and Textual Data
Kim, Kai
Tsai, Howard
Sen, Rajat
Das, Abhimanyu
Zhou, Zihao
Tanpure, Abhishek
Luo, Mathew
Yu, Rose
Artificial Intelligence
Current forecasting approaches are largely unimodal and ignore the rich textual data that often accompany the time series due to lack of well-curated multimodal benchmark dataset. In this work, we develop TimeText Corpus (TTC), a carefully curated, time-aligned text and time dataset for multimodal forecasting. Our dataset is composed of sequences of numbers and text aligned to timestamps, and includes data from two different domains: climate science and healthcare. Our data is a significant contribution to the rare selection of available multimodal datasets. We also propose the Hybrid Multi-Modal Forecaster (Hybrid-MMF), a multimodal LLM that jointly forecasts both text and time series data using shared embeddings. However, contrary to our expectations, our Hybrid-MMF model does not outperform existing baselines in our experiments. This negative result highlights the challenges inherent in multimodal forecasting. Our code and data are available at https://github.com/Rose-STL-Lab/Multimodal_ Forecasting.
title Multi-Modal Forecaster: Jointly Predicting Time Series and Textual Data
topic Artificial Intelligence
url https://arxiv.org/abs/2411.06735