Stock prediction method based on multi-source data of social media and Informer model
Yanxi Zhu · 2025
Stock market forecasting has always been an important research topic in the financial field, while traditional forecasting methods are often difficult to effectively cope with the high volatility and nonlinear characteristics of the market. In recent years, massive user sentiment and opinion data generated by social media platforms have provided new information dimensions for stock forecasting. This paper proposes a stock forecasting method based on social media multi-source data and Informer model. We combined data from Twitter, Reddit and financial news, then fed it into our modified Informer model to better predict stock movements. Experiments were conducted on the constituent stocks of the S & P 500 Index, and the results showed that: (1) compared with traditional models such as LSTM and GRU, the proposed method improved the prediction accuracy by 8.7% on average, and the MAE and RMSE decreased by 11.3% and 9.6% respectively; (2) the fusion of multi-source social media data significantly improved the model's prediction sensitivity to market emergencies; (3) Our tweaked Informer model handled long-series forecasting surprisingly well - pushing prediction timeframes about 2-3 times further than old-school methods. Beyond just adding to forecasting methods, our work offers investors something they can actually rely on when making decisions.