PriSM: Privacy-Preserving Social Media Text Processing and Analytics Framework
Sakib Shahriar, Rozita A. Dara · 2024
The rise of big data has transformed various domains by enabling data-driven insights. However, this transformation has also brought forth privacy concerns, especially in the context of social media analytics. Traditional privacy-preserving methods, such as anonymization and de-identification, are insufficient due to the complex, unstructured nature of social media texts. Consequently, this paper presents a comprehensive privacy-preserving framework and a dedicated pipeline for privacy-aware social media text analytics. The proposed framework systematically identifies and categorizes six key privacy risks in social media data. Complementing the framework, the privacy-preserving pipeline integrates privacy considerations into every data processing stage, from collection to model development. To test the effectiveness of this pipeline, we implemented a case study on a Facebook dataset, which showed a 13.64% reduction in authorship detection accuracy while maintaining the utility of the data for sentiment analysis and topic modeling.