Abstract:Modern time-series forecasting models often fail to make full use of rich unstructured information about the time series themselves. This lack of proper conditioning can lead to obvious model failures; for example, models may be unaware of the details of a particular product, and hence fail to anticipate seasonal surges in customer demand in the lead up to major exogenous events like holidays for clearly relevant products. To address this shortcoming, this paper introduces a novel forecast post-processor -- which we call LLMForecaster -- that fine-tunes large language models (LLMs) to incorporate unstructured semantic and contextual information and historical data to improve the forecasts from an existing demand forecasting pipeline. In an industry-scale retail application, we demonstrate that our technique yields statistically significantly forecast improvements across several sets of products subject to holiday-driven demand surges.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that modern time - series prediction models fail to fully utilize the rich unstructured information about the time series itself, resulting in obvious model failures when predicting certain seasonal events (such as holidays). Specifically, existing prediction models may be unable to be aware of the details of specific products, and thus cannot accurately predict the surge in customer demand before and after major exogenous events (such as holidays), especially in the retail environment. ### Core of the problem 1. **Limitations of existing models**: - Existing time - series prediction models can usually only effectively integrate numerical or categorical exogenous features, while ignoring a large amount of descriptive and contextual information in the form of unstructured text, such as product descriptions, customer reviews, etc. - This unstructured text information is difficult to be directly used for model training, so it is ignored or simply processed. 2. **Challenges in seasonal demand prediction**: - Many new products have only limited historical sales data, and these data may contain noise (such as sudden sales peaks or out - of - stock situations), making prediction based on historical data difficult. - The demand surges for certain seasonal events (such as Mother's Day, Easter) are limited to a few products, the overall customer traffic is not significant, and the product discount frequency is low, which increases the difficulty of prediction. 3. **Need for human intervention**: - Currently, in order to make up for the deficiencies of model prediction, it is necessary to rely on human analysts to identify products related to upcoming events based on experience and knowledge, and adjust the prediction accordingly. ### Solution To solve these problems, the paper proposes a new post - prediction - processing method - LLMForecaster. This method fine - tunes large language models (LLMs) to combine unstructured semantic and contextual information with historical data, thereby improving the output of the existing demand prediction pipeline. Specifically: - **Introducing LLMForecaster**: Use the fine - tuned LLMs to combine unstructured text information (such as product descriptions) and numerical features (such as price) to generate an adjustment factor for correcting the prediction results of existing models. - **Improving prediction accuracy**: In this way, LLMForecaster can systematically improve the performance of existing models in predicting holiday - related demand surges, especially in the retail environment, and improve the accuracy of product - level demand prediction. ### Experimental verification The paper verifies the effectiveness of LLMForecaster through a series of experiments, especially showing significant improvement in product demand prediction during multiple important holidays (such as Halloween, Easter, Mother's Day, Father's Day, Valentine's Day). The experimental results show that LLMForecaster can not only more accurately capture these seasonal demand surges, but also significantly reduce the risk of inventory shortages and improve customer satisfaction. ### Summary In conclusion, this paper aims to improve time - series prediction models by introducing LLMForecaster and using unstructured text information, especially making significant progress in dealing with seasonal demand surges. This method provides a better tool for the retail industry to manage seasonal fluctuations and optimize operational efficiency.

LLMForecaster: Improving Seasonal Event Forecasts with Unstructured Textual Data

From News to Forecast: Integrating Event Analysis in LLM-Based Time Series Forecasting with Reflection

Effective LSTMs with Seasonal-Trend Decomposition and Adaptive Learning and Niching-Based Backtracking Search Algorithm for Time Series Forecasting

A Comprehensive Evaluation of Large Language Models on Temporal Event Forecasting

Approaching Human-Level Forecasting with Language Models

Time-LLM: Time Series Forecasting by Reprogramming Large Language Models

Time Series Forecasting with LLMs: Understanding and Enhancing Model Capabilities

Quantifying Qualitative Insights: Leveraging LLMs to Market Predict

LLM4TS: Aligning Pre-Trained LLMs as Data-Efficient Time-Series Forecasters

Humans vs Large Language Models: Judgmental Forecasting in an Era of Advanced AI

Reprogramming Foundational Large Language Models(LLMs) for Enterprise Adoption for Spatio-Temporal Forecasting Applications: Unveiling a New Era in Copilot-Guided Cross-Modal Time Series Representation Learning

Text2TimeSeries: Enhancing Financial Forecasting through Time Series Prediction Updates with Event-Driven Insights from Large Language Models

MM-Forecast: A Multimodal Approach to Temporal Event Forecasting with Large Language Models

LeMoLE: LLM-Enhanced Mixture of Linear Experts for Time Series Forecasting

Temporal Data Meets LLM -- Explainable Financial Time Series Forecasting

Can Language Models Use Forecasting Strategies?

CALF: Aligning LLMs for Time Series Forecasting via Cross-modal Fine-Tuning

CLLMate: A Multimodal LLM for Weather and Climate Events Forecasting

An Evaluation of Standard Statistical Models and LLMs on Time Series Forecasting

News-based forecasts of macroeconomic indicators: A semantic path model for interpretable predictions

Understanding the Role of Textual Prompts in LLM for Time Series Forecasting: an Adapter View