Online Text Risk Assessment Based on the BERT-CNN-BiLSTM Model

With the rapid development of social networks and online platforms, the hidden harmful information in network texts poses significant threats to public opinion security and user mental health. To achieve efficient risk identification and assessment of online texts, this paper proposes a deep learning model integrating Bidirectional Encoder Representation from Transformer (BERT), Convolutional Neural Network (CNN), and Bidirectional Long Short-Term Memory (BiLSTM) for multidimensional semantic modeling and risk level classification. First, the BERT model is employed to extract context-sensitive text representations, enhancing semantic comprehension capabilities. Subsequently, CNN captures local textual features, while BiLSTM further explores temporal information and deep semantic structures. Experimental results on the constructed network text risk dataset demonstrate that the proposed BERT-CNN-BiLSTM model achieves superior accuracy and robustness in textual risk identification tasks. This methodology can provide technical support for practical applications such as online public opinion monitoring, content moderation, and risk early warning systems.