[en] The vision of the Internet of Things (IoT) promises novel, intelligent applications to improve services across all industries and domains. Efficient data and service discovery are crucial to unfold the potential value of cross-domain IoT applications. Today, the Web is the primary enabler for integrating data from distributed networks, with more and more sensors and IoT gateways connected to the Web. However, semantic data models, standards and vocabularies used by IoT vendors and service providers are highly heterogeneous, which makes data discovery and integration a challenging task.
Industrial and academic research initiatives increasingly rely on Semantic Web technologies to tackle this challenge. Ongoing research efforts emphasize the development of formal ontologies for the description of Things, sensor networks, IoT services and domain-dependent observations to annotate and link data on the Web. Within this context, there is a research gap in investigating and proposing ontology recommendation approaches that foster the reuse of most suitable ontologies relevant to semantically annotate IoT data sources. Improved ontology reuse in the IoT enhances semantic interoperability and thus facilitates the development of more intelligent and context-aware systems. In this dissertation, we show that ontology recommendation can form a key building block to achieve this consensus in the IoT. In particular, we consider large-scale IoT systems, also referred to as IoT ecosystems, in which a wide range of stakeholders and service providers have to cooperate. In such ecosystems, semantic interoperability can only be efficiently achieved when a high degree of consensus on relevant ontologies among data providers and consumers exists.
This dissertation includes the following contributions. First, we conceptualize the task of ontology recommendation and evaluate existing approaches with regard to IoT ecosystem requirements. We identify several limitations in ontology recommendation, especially concerning the IoT, which motivates the main focus on ontology ranking in this dissertation. Second, we subsequently propose a novel approach to ontology ranking that offers a fairer scoring of ontologies if their popularity is unknown and thus helps in providing a better recommendation in the current state of the IoT. We employ a `learning to rank' approach to show that qualitative ranking features can improve the ranking performance and potentially substitute an explicit popularity feature. Third, we propose a novel ontology ranking evaluation benchmark to address the lack of comparison studies for ontology ranking approaches as a general issue in the Semantic Web. We develop a large, representative evaluation dataset that we derive from the collected user click logs of the Linked Open Vocabularies (LOV) platform. It is the first dataset of its kind that is capable of comparing learned ontology ranking models as proposed in the literature under real-world constraints. Fourth, we present an IoT ecosystem application to support data providers in semantically annotating IoT data streams with integrated ontology term recommendation and perform an evaluation based on a smart parking use case.
In summary, this dissertation presents the advancements of the state-of-the-art in the design of ontology recommendation and its role for establishing and maintaining semantic interoperability in highly heterogeneous and evolving ecosystems of inter-related IoT services. Our experiments show that ontology ranking features that are well designed with regard to the underlying ontology collection and respective user behavior can significantly improve the ranking quality and, thus, the overall recommendation capabilities of related tools.