아주대학교

목차

I. Introduction 1
1.1 Motivation and Objectives 1
1.2 Research Contributions 7
1.3 Organization of the Dissertation 8
II. Background 9
2.1 Language Modeling 9
2.2 Personally Identifiable Information 11
2.3 Training data availability issue 12
2.4 Impact of data redundancy on privacy risks 13
2.4.1 Experiment and Result 14
2.5 Privacy impact on PII Linkability 15
2.5.1 Our proposed Knowledge Probing method 15
2.5.2 Experiment and Result 17
2.6 Related works 18
III. PrivGraph: Modeling Personal Privacy Information using Knowledge Graph 21
3.1 Definition of PII data category 21
3.2 Specifying data category 23
3.3 Definition of PrivGraph 23
3.4 Sensitivity Level Factor 25
IV. PrivGrapher: an NLP pipeline to build PrivGraph 27
4.1 NLP-based Named Entity Recognition 30
4.2 Data type Classification 32
4.3 Individual Recognition 32
4.4 Relationship Annotation 33
4.5 Graph Merging 36
4.6 PrivGraph Construction 37
V. PrivGrapher Evaluation 39
5.1 Dataset 39
5.2 Baseline Models 41
5.3 Result 43
VI. PrivGraph Application 44
6.1 Modeling PII from different sources 44
6.1.1 Experiment Result 44
6.2 Long textual data analysis 45
6.2.1 Experiment result 45
6.3 PrivGraph based privacy knowledge probing 47
6.3.1 Experiment result 50
6.4 Discussion of applying PrivGraph in the real world 50
VII. Conclusion and Future work 54
7.1 Conclusion 54
7.2 Future work 55
References 56