Current information technologies allow people to acquire personal data related to their health, lifestyle, behavior, and activities, often using wearable and mobile devices. Personal data management technologies have emerged recently, in order to cope with the requirements of this type of data, ranging from personal clouds to self-storage solutions. Pryv.io is a comprehensive solution for managing this particularly sensible type of data streams, focusing both on data privacy and decentralization. In this paper, we describe SemPryv, a system aiming at providing a semantization mechanism for enriching personal data streams with standardized specialized vocabularies from third-party providers. It relies on third providers of semantic concepts, and includes rule-based mechanisms for facilitating the semantization process. A full implementation of SemPryv has been produced, pluggable to the existing Pryv.io platform, showing the feasibility of the approach.
The increasing amount of generated personal data allows for the development of personalized applications in different domains, usually related to health, lifestyle, or everyday activities. These often rely on different sources and acquisition modalities, including wearable devices, sensors, domotic technologies, or self-reporting methods. In this context, it is essential to provide data privacy guarantees, in order to avoid unintended access or disclosure. This difficulty to address this challenge is further increased due to the streaming nature of many of these datasets, which require infrastructure designed to manage high-volume and high-velocity information flows.
Pryv.io is a privacy-centric middleware, used as a robust data management
foundation to develop risk-controlled mHealth, eHealth, and InsurTech
applications with confidence and in respect to IT and regulatory requirements. Pryv.io
is built based on two key pillars: decentralization and privacy. Unlike traditional
centralized solutions, Pryv.io stores each data account separately and
independently, making it possible to be even deployed on its own server [
This paper describes the SemPryv subsystem for stream data enrichment3.
The goal of SemPryv is to provide semantization capabilities for the Pryv.io
middleware, such that it can automatically propose semantic concepts,
associated to the heterogeneous data streams managed by the platform. The data
semantization makes it possible to enhance the data model, currently conformed
by typed events. Associating high-level ontology concepts to the stream events
enables new types of search and discovery functionalities in the middleware,
which were not possible up to now. Also, it provides the means to link the Pryv
datasets with existing standards and models used widely for cataloguing data
in the health sector. In particular, the use of standards, such as HL7 FHIR [
The Pryv.io middleware is used to manage large and diverse streams of data
coming from external platforms, wearable devices, and health record systems.
These streams are organized through identifiers and tags that are later used for
searching and querying. While it is technically possible to export and make the
Pryv datasets available to external applications in different formats, standards
such as HL7 FHIR [
The architecture of SemPryv is depicted in Figure 2. SemPryv has two main
components: a back-end that exposes the core services as a REST API, and a UI
for end-users and experts. Besides the proxy capabilities mentioned before, the
SemPryv back-end can connect to a series of providers for semantic vocabularies
and ontologies. These may include existing APIs such as BioPortal [
The architecture presented previously describes how the different components of the system interact with each other. Concerning the semantization process itself, the SemPryv module is flexible enough to adapt to different types of situations.
For users and data integrators, the SemPryv UI (Figure 4) exposes the proposed semantics, queried from the 3rd party providers (e.g. BioPortal). Then, these suggestions can optionally be confirmed by an administrator before being consolidated into its corresponding Pryv instance. This semi-automatic semantiza- Fig. 4. SemPryv UI: suggested annotion makes it possible to have full control tations obtained from the BioPortal over the type of semantics to be assigned. provider: SNOMED-CT terms. As an example, a body-weight stream in the Pryv.io middleware can be modeled as the weight of an individual according to SNOMED-CT, codified as: SNOMED-CT:27113001. Notice that multiple annotations can be attached to a given stream, and that these annotations can be inherited recursively by sub-streams and events inside of its hierarchy. Furthermore, other custom 3rd party ontology/vocabulary providers can be configured in order to feed the system. In addition, SemPryv includes the possibility of using predefined rules expressed in its knowledge graph. These rules can be modified by administrators, and essentially allow the definition of close terms from different ontologies. For instance in the following example, the knowledge graph matches Pryv temperature streams to a SNOMET-CT code identified as: snomed-ct:386725007. Similarly the same is done for mass. Then, the system also allows to match these rules to certain stream paths, defined using regular expressions. "@graph": [{ "@id": "pryv:temperature", "@type": "skos:Concept", "skos:notation": "note/txt", "skos:closeMatch": "snomed-ct:386725007", }, { {
Listing 1.1. Predefined rules mapping pryv concepts to SNOMED-CT.
In this paper we have described SemPryv, a system that allows the semantic enrichment of personal data streams, set up in a fully distributed environment. The proposed approach has been fully implemented, comprising not only the semantization but also (i) its integration with external providers such as BioPortal, (ii) the implementation of an interoperability bridge through HL7 FHIR, and (iii) a rule-based automated suggestion feature. The system is currently deployed, showcasing the use of semantics in real-life scenarios and on integrated with a commerical solution. The SemPryv approach relies on two main principles. First, on the reuse of consolidated vocabularies, ontologies, and taxonomies that are standardized and widely used in the domains of application. This is the case for well-known standards (eg. SNOMED-CT, LOINC; UCUM, RxNorm) which have been curated to enable interoperability among applications. Second, SemPryv uses different, but complementary approaches for proposing semantics for a given dataset, depending on: the available data, metadata, and previous inferences. For the next iteration of the SemPryv module, we will further enhance the rule-inferencing approach for establishing suggestions of semantic metadata, in cases where a bootstrapping process is required. Once a critical mass of data is acquired, SemPryv will rely on incremental machine learning techniques to correlate previously annotated datasets with new incoming streams. The prototype is publicly available, as referenced earlier, and in the future an Open Source approach will be considered, given the potential reuse opportunities.
Acknowledgements Supported by the InnoSuisse grant 26547.1 SemPryv project.