科研成果详情

With the rapid global expansion and popularity of short video platforms, the identification and filtering of non-compliant content have become crucial tasks to ensure the health of the online environment. Traditional manual review methods face challenges of low efficiency and poor consistency, urgently requiring an efficient, automated solution. This study proposes a deep learning model based on BERT and TextCNN, aimed at automatically identifying non-compliant information within Chinese text content. By integrating the deep semantic understanding capabilities of the BERT model with the local feature extraction advantages of TextCNN, we designed and implemented an effective model for non-compliant information identification. This research first conducted a detailed preprocessing and analysis of the dataset, including exploring word length distribution, analyzing the proportion of compliant and non-compliant information, and visualizing key vocabulary through word cloud graphics. Subsequently, we trained and tested the model, which achieved an accuracy of 93.61% on the identification task, demonstrating good balance across precision, recall, and F1 score metrics, with an AUC value of 0.96. This indicates the model's high accuracy and reliability in distinguishing between compliant and non-compliant information. The outcomes of this study not only provide an effective automatic identification tool for short video platforms but also offer a new research perspective and practical evidence for the application of deep learning in text analysis.