The efficient market hypothesis in traditional financial theory struggles to explain the short-term irrational fluctuations in the A-share market, where investor sentiment fluctuations often serve as the core driver of abnormal stock price movements. Traditional sentiment measurement methods suffer from limitations such as lag, high misjudgment rates, and the inability to distinguish confounding factors. To more accurately explore the dynamic correlation between investor sentiment and stock price fluctuations, this paper proposes a sentiment analysis framework based on large language models (LLMs). By constructing continuous sentiment scoring factors and integrating them with a long short-term memory (LSTM) deep learning model, we analyze the correlation between investor sentiment and stock price fluctuations. Empirical results indicate that sentiment factors based on large language models can generate an annualized excess return of 9.3% in the CSI 500 index domain. The LSTM stock price prediction model incorporating sentiment features achieves a mean absolute percentage error (MAPE) as low as 2.72%, significantly outperforming traditional models. Through this analysis, we aim to provide quantitative references for optimizing investment decisions and preventing market risks.