AI-BASED PRIVACY PRESERVING TEXT TRANSFORMATION

Abstract

he rapid growth of digital platforms has significantly increased the risk of exposing sensitive personal information present in textual data. Protecting such data while maintaining its usability has become a critical challenge in modern data processing systems. This project, “AI-Based PrivacyPreserving Text Transformation,” presents an intelligent and automated solution for safeguarding sensitive information without compromising the original meaning and readability of the text. The proposed system leverages advanced Natural Language Processing (NLP) techniques and machine learning tools such as spaCy and Microsoft Presidio to identify Personally Identifiable Information (PII), including names, phone numbers, email addresses, and other confidential data. Once detected, the system applies privacypreserving techniques like masking, anonymization, and placeholder substitution to secure the information while preserving contextual integrity. Additionally, the system supports multiformat data processing by integrating Optical Character Recognition (OCR) technologies such as PyTesseract and PDF2Image, enabling text extraction from images and scanned documents. To ensure data security, the original sensitive information is encrypted using AES-based cryptographic methods and stored securely, allowing controlled access and restoration only by authorized users through a rolebased system. The architecture of the system, as illustrated in the system design diagram on page 16, follows a multi-layered approach involving preprocessing, analysis, transformation, validation, and secure storage, ensuring scalability and efficiency. The implementation demonstrates high accuracy in detecting sensitive entities and reliable restoration of original data, achieving a balance between data privacy and usability.