DepInferAttack: Framework for Membership Inference Attack in Depression Dataset

Online social networks (OSNs), such as Reddit, capture a variety of human interactions, including sensitive discussions about depression. The study uses posts related to depression on Reddit, which gives an authentic view of how individuals express their feelings and experiences. Emotional-rich and highly disparate content makes this type of data difficult for machine learning models to process. Existing literature on membership inference attacks (MIAs) focusses on simple structured data and does not consider emotional aspects in these datasets. Our research addresses this gap by investigating MIAs with respect to emotionality, showing potential privacy concerns when using sensitive information. This paper introduces DepInferAttack a new framework designed specifically for performing MIAs over the OSN platform's depression-related content. Three different classifiers are employed in the target model: random forest, support vector machine, and neural network. A large-language model is used to generate synthetic datasets for the shadow model. The CNN architecture is employed by the shadow model, while the attack model uses a random forest classifier. The results show that DepInfer Attack framework is very effective at MIAs as it achieves a high accuracy of 94% to correctly classify whether the data point is a member of the target model's training dataset or not.