PrivacyChat: Utilizing Large Language Model for Fine-Grained Information Extraction over Privacy Policies

Privacy policies play a crucial role in upholding the privacy rights of users and fostering trust between organizations and their users. By clearly understanding the terms and conditions of a privacy policy, individuals can make well-informed choices about disclosing their personal information and understand how the concerned entity will manage their data. Following the introduction of the General Data Protection Regulation, these policies have become more extensive and intricate. This creates a challenge for users in terms of understanding and finding specific information in the policy. Today, through prompt-based methods, we can extract specific data from extensive text documents using large language models (LLMs), thus eliminating the need for training or fine-tuning models. In this study, we explore a prompt-based approach to extract information concerning personal data from privacy policies using a large language model, GPT-3.5. In this preliminary study, we assess the performance of GPT-3.5 on such a fine-grained extraction task through varied metrics and its capability to address previous computational challenges. The prompt structure can be adapted for other LLMs, and a similar approach can be employed for various information extraction tasks over privacy policies. The data and code are available at our GitHub repository . .