Time Series Large Language Model for Recommendation System

In recent years, Large Language Models (LLMs) have demonstrated significant performance in diverse domains, leading to increased interest in their application to recommendation systems. This integration introduces novel approaches for advancing recommendation technologies. Nevertheless, challenges remain: misalignment between LLM capabilities and specific requirements of recommendation tasks, along with inadequate exploration of intricate data correlations by LLMs, can compromise recommendation quality. Addressing these challenges, we propose a novel approach involving the finetuning of LLMs on recommendation-specific datasets to enhance alignment and introduce the Temporal Integration for Sequential Large Language Model (TisLLM) framework. TisLLM aims to deepen the extraction of latent associations within recommendation task data. It achieves this by segmenting recommendation data using temporal sliding windows, which are then aggregated before being used to fine-tune LLMs for improved compatibility with recommendation tasks. Our empirical evaluations on standard datasets have yielded favorable outcomes, affirming the efficacy of the TisLLM framework.