Does media reports on swine diseases improve the prediction accuracy of pork Price fluctuations? Evidence from China

BACKGROUND Understanding and accurately forecasting pork price volatility are critically important for the stable development of the pig industry and for designing targeted regulatory policies. In the era of a highly developed internet, media reports on swine diseases have emerged as a significant factor influencing price fluctuations, yet their role in improving prediction accuracy has not been rigorously quantified. Based on monthly data from China spanning 2012–2023, the present study constructed a hybrid forecasting framework integrating media report indicators (exposure and sentiment) with Crested Porcupine Optimizer (CPO)‐feature mode decomposition (FMD) secondary decomposition and several prediction models to assess whether media reports on swine diseases improve the prediction accuracy of pork price fluctuations. RESULTS The proposed CPO‐convolutional neural network (CNN)‐long short‐term memory (LSTM)‐attention model delivered the best predictive performance among all compared models, reducing mean absolute error (MAE) by 29.2% compared to CNN‐LSTM‐attention and by 48.5% compared to random forest and support vector machine, at the same time as increasing R 2 from approximately 0.64 to 0.920. Applying secondary decomposition to the price series yielded markedly better prediction results than single decomposition. The residual term from the initial decomposition preserved structural information relevant to price fluctuations; a second decomposition increased the signal‐to‐noise ratio of input features, resulting in a 41.4% reduction in MAE, a 41.7% improvement in R 2 and a decrease in mean squared error from 0.002 to 0.001. Adding media report indicators further increased prediction accuracy. Both exposure and sentiment contributed to prediction accuracy, with sentiment exerting a substantially greater marginal effect than exposure. The full model achieved an R 2 of 0.962 and an MAE of 0.017. Removing exposure caused R 2 to decrease to 0.885 and MAE to increase to 0.020. Removing sentiment led to a larger decline ( R 2 = 0.793, MAE = 0.030). When both media indicators were removed, the model performed worst ( R 2 = 0.654, MAE = 0.037). CONCLUSION Media reports on swine diseases significantly improve the prediction accuracy of pork price fluctuations in China, with sentiment playing a more influential role than exposure. The hybrid framework integrating secondary decomposition and media report indicators is highly effective for pork price forecasting. These findings provide empirical evidence for incorporating media‐based public opinion into agricultural price prediction models and offer practical insights for policy design and industry stabilization.