Stock market is usually characterized by drastic fluctuations, long volatility cycles, and emotional changes. This paper focuses on exploring how to utilize complex networks and sentiment analysis techniques to improve the prediction accuracy of stock movements. We take the semiconductor industry chain stock market as the research object. First, based on complex network theory, we construct a complex network of China’s semiconductor industry chain and extract key network factors. Second, the semantic understanding ability of the large language model is applied to investor comment sentiment analysis, and a multi-dimensional stock sentiment factor is constructed by fine-tuning the GPT large model and combining it with the KNN classifier to recognize investor comment sentiment. Finally, the sentiment factor is combined with the network factors to construct a stock price prediction model using the LSTM algorithm, which is used to predict future stock movements. The experimental results show that the investor sentiment recognition method based on GPT-KNN can obtain an accuracy of 77.59%, which is better than the comparison algorithm, proving the significant advantage of the GPT model in text semantic feature extraction. The average accuracy of the stock trend prediction method integrating multiple factors reaches 69.62%, which is significantly better than the benchmark. Meanwhile, the experimental results of feature combination show that the accuracy is improved by 8.46% and 3.1% by the integration of network factors and sentiment factors, respectively.