Gespeichert in:
| Hauptverfasser: | , |
|---|---|
| Format: | Recurso digital |
| Sprache: | |
| Veröffentlicht: |
Zenodo
2025
|
| Online-Zugang: | https://doi.org/10.5281/zenodo.17693446 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Inhaltsangabe:
- <p>Context: Processes that aim to extract value from stored data are gaining prominence. Among various types of unstructured data, textual data constitutes a significant proportion of the information produced in real-world settings. Ethical considerations and data protection laws have increased the pressure over the privacy of sensitive content. The disclosure risks associated with textual data, considering differential privacy, are influenced by the rarity and the similarity of texts within a dataset. Rare texts can increase the likelihood of re-identification. Artificial Intelligence (AI) and Machine Learning (ML) have growing demand for data and side by side with statics and classic natural language processing techniques, those techniques are increasingly being explored for implementing privacy-preserving mechanisms, offering technical solutions to mitigate privacy risks. Goal: The objective was uncover state of art techniques for the privacy-preserving when processing of text data for allow to employ techniques to protect privacy in unstructured data, specifically textual, and when implementing text similarity techiques. Method: To achieve this goal, state-of-the-art privacy preservation techniques were researched, in a literature review, and the study proposed the application of selected techniques. The concepts of differential privacy, vector databases, text similarity and rare events were taken into account in the proposed methodology and case study, along with the use of multi-agent Artificial Intelligence (AI) systems and Large Language Models (LLMs). Results: A key contribution of this study was to identify the state of art techniques for the privacy-preserving that are applied in textual data analysis and text similarity, including as how Data Science, LLM and Agent-Based AI techniques are used to implement privacy-preserving mechanisms and the techniques that are employed for semantic similarity and rare events detection in text domains. And also, a purposed application in case study for the use of that knowledge. Conclusion: This study offers both a structured synthesis of existing studies through a Sytematic Literature Review (SLR) and a practical perspective via a case study, highlighting privacy-preserving techniques in text analysis. It highlights the possibility of using semantic similarity methods and vector-based representations in identifying rare events in contexts under privacy constraints. The integration of LLMs and AI agents reveals promising but in other hand there are specific challenges and complexity for privacy-aware processing, particularly in areas like public security. This study provided an overview of the implementation and practical use, applied in a case study, of semantic similarity techniques between texts, which were revealed in SLR to have a strong and mature presence in the literature. Given the scarcity of similar approaches in the surveyed literature, this work addresses a a contribution to help minimize this notable gap and try to contribute for future research focused on reconciling AI methods with ethical, privacy-preserving applications.</p>