Energy Policy Text Sentiment Analysis Using a Dual-Channel Approach with BERT Embeddings and KAN-BiLSTM

The task of sentiment analysis in energy policy texts presents unique challenges. Most existing methods rely on general word vector models and fail to fully consider the semantic and emotional characteristics specific to the domain. To more accurately capture the sentiment tendencies in energy policy texts, this paper proposes a new model based on the combination of BERT and KAN-BiLSTM with an attention mechanism - BKBA (BERT-KAN-BiLSTM-Attention). The model processes text through two channels: using the BERT model for word embedding, one channel employs KAN to extract text features, enhancing domain knowledge understanding; the other channel uses a BiLSTM network to capture contextual information and global dependencies, ensuring effective handling of long-distance dependencies. The attention mechanism is introduced to focus the model on important features and promote deep interaction between features, better capturing the textual sentiment trends. Experiments show that the BKBA model performs excellently in precision, recall, and F1 score, validating its superiority in the task of sentiment analysis for energy policy texts. Compared with other models, the effectiveness and advantages of the BKBA model are further demonstrated.