This paper focuses on extracting relevant information from unstructured data, specifically analyzing text shared by users on Twitter. The goal is to build a comprehensive knowledge graph by extracting implicit personal information from tweets, including interests, activities, events, family, health, relationships, and professional information. The extracted information is used to instantiate a digital twin and develop a personalized alert system to protect users from threats, such as social engineering or doxing. The paper evaluates the effectiveness of state-of-the-art large language models, such as GPT-4, for extracting relevant triples from tweets. The study also explores the notion of digital twins in the context of cyber threats and presents related work in information extraction. The approach includes data collection, multi-label classification, relational triple extraction, and evaluation of the results. The dataset used is from Twitter, and the study analyzes the challenges posed by user-generated data. The results show the accuracy of the extracted triples and the personal characteristics that can be identified from tweets for the development of the Digital Twin. The results contribute to the ADRIAN research project, which focuses on machine learning-based methods for detecting potential threats to people's privacy.    «

This paper focuses on extracting relevant information from unstructured data, specifically analyzing text shared by users on Twitter. The goal is to build a comprehensive knowledge graph by extracting implicit personal information from tweets, including interests, activities, events, family, health, relationships, and professional information. The extracted information is used to instantiate a digital twin and develop a personalized alert system to protect users from threats, such as social engi...    »