Deep analysis of ideological and political education (IPE) data is vital for enhancing pedagogical precision. Addressing limitations of current methods—particularly in comprehending complex semantics, integrating domain knowledge, and evaluating civic-political (Si-Pol) characteristics—this study proposes an improved deep learning algorithm. Our framework augments semantic understanding by structuring external knowledge, employs a hierarchical attention mechanism to capture key textual information, and incorporates Si-Pol features into the loss function design. Evaluated on the COAE dataset, the model achieved macro-averaged F1 scores of 87.3% and 83.6% on the critical tasks of sentiment orientation recognition and values consistency assessment, respectively. This represents a significant improvement of 3.7% to 8.2% over baseline models (TextCNN, BiLSTM-Attention, base Transformer). The research provides a novel pathway for more accurate and interpretable analysis of IPE effectiveness.