Network Embedding Technology Based on Breadth Learning for Information Extraction and Review in Social Media

Abstract

Social media platforms have brought tremendous changes to the way society communicates and transmits information, but there are also some extreme individuals who post sensitive information such as false, malicious attacks, terrorism or negative comments. To address this rapidly evolving network security challenge, this study introduces breadth learning and network embedding techniques from machine learning to establish a new method for information extraction and auditing. This method adopts a meta path random walk form to extract and filter effective information from multiple heterogeneous social networks, and then proposes a network embedding technique based on breadth learning to transform each node into a low dimensional feature space. Finally, a fusion function is designed to quantify the correlation between sensitive users on social platforms, accurately identifying sensitive information publishers from multiple social platforms and effectively locking in sensitive information. As a result, a sensitive information extraction and review model based on breadth learning and network embedding is constructed. The results showed that the research model had a testing accuracy of 97.3% in mixed datasets on social platforms, which was better than other sensitive information review models, and performed better on social platforms with larger data volumes. The research results have significant implications for the network security governance of open large-scale social media platforms.