知识图谱增强的残差集成网络舆情热度预测方法OA

To address the challenge of integrating multi-source heterogeneous data and capturing deep semantic correlations in public opinion prediction,this paper proposed a knowledge graph-enhanced auto-residual regression ensemble(KG-ARRE)framework.The KG-ARRE framework extracted daily features from microblog data,built a subject-object interaction know-ledge graph using textual semantic analysis,and applied an auto-regressive integrated moving average(ARIMA)model to ob-tain residuals from the popularity time series.Then it combined the multi-source features by temporal convolutional network(TCN)for modeling.The framework also used multiple randomly initialized sub-models and an error-weighted ensemble me-chanism to improve prediction stability.Experiments on real-event datasets show that KG-ARRE achieves a mean absolute per-centage error(MAPE)of 4.53％,which is 63.1％lower than that of the baseline TCN(MAPE=12.28％).When the knowledge graph and ensemble modules are transferred to models such as BiLSTM,the MAPE decreases by 2.36～7.05 per-centage points.These results demonstrate that KG-ARRE enhances prediction accuracy and generalization.