Sports social media influence prediction model with temporal transformer and causal reasoning

Predicting the spread of information in sports social media remains challenging due to the complex interplay between dynamic propagation processes and confounding contextual factors.To move beyond purely correlation-driven approaches, this paper proposes a novel twin-stream temporal transformer architecture integrated with a causal inference module.This model concurrently encodes sequences of match events and social media propagation states, while a counterfactual reasoning component adjusts for potential confounders.Evaluated on a real-world dataset from professional soccer, our framework achieves a root mean square error of 0.412, a mean absolute error of 0.298, and an area under the curve of 0.891, outperforming existing benchmarks by 6.3% to 14.7% across key metrics.The model not only enhances predictive accuracy but also quantifies the causal effect of specific match events, offering both robust forecasting and explainable insights into the drivers of social media engagement.