This paper proposes a Dual-channel model named DCFI (identifying malicious comments by dual-channel combined Multi-Dimensional FI) to address the problem of identifying malicious comments. The model combines the Graph Convolutional Network (GCN) and Bidirectional Long Short-Term Memory (BiLSTM) to extract different textual features, enabling a more comprehensive representation of the text’s meaning. The proposed Multi-Dimensional FI refines and fuses the extracted features, followed by a fully connected layer and softmax classification to achieve the identification of malicious comments. Experimental results demonstrate that the DCFI model outperforms mainstream classification models in terms of malicious comment identification. It improves classification accuracy by 1.06% to 2.89%. This approach effectively extracts textual features specific to malicious comments, enabling accurate identification.