Artificial intelligence systems such as Large Language Models (LLM) derive their knowledge from large datasets. Systems like ChatGPT therefore rely on shared data to train on. For companies, releasing data to the public domain requires anonymization as soon as an individual is identifiable. While there are several privacy models that guarantee a certain level of distortion applied to a dataset, to mitigate re-identification, e.g. with k-anonymity, the required level is generally defined by the data processor. We propose the idea of combining individual privacy levels defined by the data subjects themselves with a privacy language, such as the Layered Privacy Language (LPL) [10], to get a more fine-grained understanding of the effectively required privacy level. Queries that target subsets of the dataset to be released can only benefit from lower privacy requirements set by data subjects, as these response subsets may do not contain users with high privacy requirements, which can then lead to more utility. By analyzing the results of different queries to a privacy-aware data-transforming database system, we demonstrate the characteristics required for this assumption to be truly effective. For a more realistic evaluation, we also consider changes in the underlying data sources.

The overweight and obesity rate is increasing for decades worldwide. Healthy nutrition is, besides education and physical activity, one of the various keys to tackle this issue. In an effort to increase the availability of digital, healthy recommendations, the scientific area of food recommendation extends its focus from the accuracy of the recommendations to beyond-accuracy goals like transparency and healthiness. To address this issue a data basis is required, which in the ideal case encompasses user-item interactions like ratings and reviews, food-related information such as recipe details, nutritional data, and in the best case additional data which describes the food items and their relations semantically. Though several recipe recommendation data sets exist, to the best of our knowledge, a holistic large-scale healthiness-aware and connected data sets have not been made available yet. The lack of such data could partially explain the poor popularity of the topic of healthy food recommendation when compared to the domain of movie recommendation. In this paper, we show that taking into account only user-item interactions is not sufficient for a recommendation. To close this gap, we propose a connected data set called HUMMUS (Health-aware User-centered recoMMendation and argUment-enabling data Set) collected from Food.com containing multiple features including rich nutrient information, text reviews, and ratings, enriched by the authors with extra features such as Nutri-scores and connections to semantic data like the FoodKG and the FoodOn ontology. We hope that these data will contribute to the healthy food recommendation domain.

The transformation of big data to the cloud requires us to reconsider trust. Trust in all parties involved in the data management, the infrastructure as well as all groups with an access interest. A common way to mitigate the risk of the identification of individuals in case of privacy breaches is anonymization, which consequently also leads to information loss. Depending on the assumed level of confidence, a data processor can control the risk for privacy breaches in changing the point where anonymization gets applied. We examined anonymization points in data warehouse scenarios to evaluate their effects on utility and re-identification risk. Our evaluation showed that data quality can differ up to 4,80% while the re-identification risk is reduced by up to 16,82 %. With still improved quality, the re-identification risk differs up to 53,49 % in another configuration.

Protecting privacy of individuals is a basic right, which has to be considered in our data-centered society in which new technologies emerge rapidly. To preserve the privacy of individuals de-identifying technologies have been developed including pseudonymization, personal privacy anonymization, and privacy models. Each having several variations with different properties and contexts which poses the challenge for the proper selection and application of de-identification methods. We tackle this challenge proposing a policy-based de-identification test framework for a systematic approach to experimenting and evaluation of various combinations of methods and their interplay. Evaluation of the experimental results regarding performance and utility is considered within the framework. We propose a domain-specific language, expressing the required complex configuration options, including data-set, policy generator, and various de-identification methods.

In out data-driven society more and more personal and sensitive data is processed and stored making it virtually impossible for end-users to comprehend what happens to their data. Although in health care strict regulations for the processing of personal data already existed, the General Data Protection Regulation (GDPR) provides a EU-wide regulation. Despite these regulations, it still exists a lack of transparency due to high complexity and missing details of privacy policies. The lack of transparency increases when various services are integrated sharing their data and forming virtual data marketplaces with various stakeholders. We argue for the strategic usage of privacy languages, i.e. the Layered Privacy Language (LPL), to formalize and present privacy policies transparently to users, enable consent management, and personalization of privacy requirements. Therefore, LPL policies are intended to fill the gap between the statement of privacy and its realization. Although LPL has been designed with the requirements of privacy policies considering GDPR, real-life privacy policies are required to be expressible with its vocabulary. Therefore, LPL will be validated against a meaningful real-life privacy policy example, that can reflect the future of integrated health care services to demonstrate capabilities, compliance and limitations of privacy languages.

On the 25th May 2018 the General Data Protection Regulation (GDPR) will enter into force implying new challenges to both legal and computer sciences. The Layered Privacy Language (LPL) is intended to model privacy policies to enforce policy-based, privacy-preserving processing. In this paper, we identify requirements for privacy policies based on Art. 12 - 14 of the GDPR, analyze LPL according to the derived requirements, and propose improvements for LPL accordingly.