with the development of communication equipment like network technology, the articles reported by media can convey information to public readers more timely by all kinds of channels such as news and so on. Therefore, the passages have more impact on the stock price. They may contain valuable information that reflect the change of stock’s fundamentals. In this paper, we utilize the natural language processing (nlp) in deep learning to analyze the stock related news. First, we use large language model (LLM) to map each word of text to an embedding vector. Then we build an attention-based model for stock movement prediction on China’s stock market according to the word embedding vector. A trading strategy can be obtained by the predicted result. Furthermore, we find that when financial knowledge graph is used to measure the relation between stocks, trading profit can be strengthened if we trading the related stock simultaneously. Comparing with previous method, deep learning model shows great potential for feature extraction and stock price prediction.