Research on text content detection and classification technology based on multi-source data fusion

In the era of digital content, detecting and classifying online information has become increasingly challenging due to the vast amount of data and the complex nature of content. Misinformation, spam, and other misleading content pose significant risks, making accurate content analysis a critical task. This study aims to develop a robust approach for text content detection and classification by integrating multiple data sources, including user behavior, temporal trends, and cross-platform metadata. The proposed model leverages BERT (Bidirectional Encoder Representations from Transformers), a powerful deep learning model for natural language processing, to extract context-rich text embeddings. These embeddings are fused with additional features using a feature-level fusion technique, and a deep neural network is employed for classification into categories such as "spam," "misleading," and "neutral." Experimental results demonstrate that the model outperforms traditional methods in precision, recall, and F1-score, particularly in distinguishing subtle variations in content. The model's effectiveness is validated through precision-recall curves, classification metrics, confusion matrices, and feature importance analysis, proving its high accuracy and adaptability. This approach has practical significance for automated content moderation, offering a more efficient and scalable solution for identifying harmful or misleading online content across platforms.