Résumé : This paper introduces a novel approach to enhance time series forecasting using Large Language Models (LLMs) and Generative Agents. With language as a medium, our method adaptively integrates various social events into forecasting models, aligning news content with time series fluctuations for enriched insights. Specifically, we utilize LLM-based agents to iteratively filter out irrelevant news and employ human-like reasoning and reflection to evaluate predictions. This enables our model to analyze complex events, such as unexpected incidents and shifts in social behavior, and continuously refine the selection logic of news and the robustness of the agent's output. By compiling selected news with time series data, we fine-tune the LLaMa2 pre-trained model. The results demonstrate significant improvements in forecasting accuracy and suggest a potential paradigm shift in time series forecasting by effectively harnessing unstructured news data.