：Gold futures prices encapsulate investor sentiment, and the validity of sentiment analysis significantly impacts price prediction accuracy. Existing studies primarily extract investor sentiment from news texts. However, the lagging nature, indirect expression, and filtering of emotions in news texts prevent them from serving as a direct reflection of investor sentiment. Therefore, this study proposes an integrated BiTS-GPriceEmo-LSTM approach for gold futures price prediction based on the sentiment analysis of investors in professional communities. The method is mainly divided into three stages. (1) Combining large language models with machine learning algorithms to extract investor sentiment-related comments from professional communities. (2) A CNN-Smoothing hybrid model is proposed to quantify investor sentiment in the extracted comments. This method introduces an innovative data smoothing technique to mitigate the high volatility of extracted investor sentiment. (3) Based on the sentiment analysis results, a dual-channel LSTM model is constructed to predict gold futures prices. Experimental results demonstrate that this approach improves the accuracy of gold futures price predictions by 3.9%.