Semantic technologies, and knowledge graphs (KGs) in particular, show promising potential for enhancing interoperability among diverse IoT systems, especially in the building sector. However, the extensive manual effort required for semantic modeling limits the widespread adoption of these technologies in engineering practice. This paper introduces Semantic-IoT, a framework that generates KGs from IoT platforms using the RDF Mapping Language (RML). By employing a semi-automated approach for declaring RML rules alongside fully automated KG generation, the framework reduces the manual effort associated with semantic modeling. The generated KGs comprehensively represent both system information and data interactions, thereby facilitating the development and deployment of cross-platform applications. A building automation use case is presented to demonstrate the feasibility and effectiveness of the proposed approach.
The RML Generation module is designed to semi-automatically create RML mapping rules for various IoT platforms. Initially, this module loads a dataset that represents the data models of an IoT platform. Typically, such a dataset comprises a list of virtual entities that encapsulate sensor data, actuation functions, and semantic information about the underlying system. From this dataset, information required by RML is extracted by dedicated submodules, as shown in Table 1.
The RML preprocessor first identifies unique resource types from the dataset based on the type field of each entity. For each resource type, the corresponding JSONPath to locate entities of that type (e.g., $[?(@.type==’TemperatureSensor’)]) and the IRI template for the RDF subject are generated. The extraction of interrelational information requires more complex processing as illustrated in Algorithm 1. Simply put, the RML preprocessor traverses the JSON object of each entity and identifies substructures that contain references to other entities. In this way, a structural skeleton of the specific data model is created.
Algorithm 1 Find interrelationship for a given entity 1: Input: An JSON object entity, a list of JSON objects allEntities 2: Output: A list foundRelationships for the given entity 3: 4: foundRelationships ← an empty list 5: for otherEntity in allEntities do 6: if entity.id ̸= otherEntity.id then 7: for (path, value) in TRAVERSEJSON(entity) do 8: for value = otherEntity.id do 9: record ← new record with fields {path, relatedType} 10: record.path ← path 11: record.relatedType ← otherEntity.type 12: Add record to foundRelationships 13: return foundRelationships
Subsequently, the subject and predicate terminologies are matched against the domain ontologies. This
submodule loads the serialized domain ontologies and then uses the Levenshtein algorithm to compute
string similarities between the terms of the data model and those in ontologies. Based on the highest
similarity scores, matching suggestions are generated. These suggestions, along with the previously
extracted information, are populated into a report for engineers to validate. Currently, manual intervention
is required to validate the suggested terminology matches and to specify URL patterns for accessing API
endpoints of the IoT platform, specifically when a resource type provides sensor data or supports actuation
functions. Once the report is finalized, the RML mapping rule can be generated automatically, thereby
eliminating the need for manual handling of RML syntax. In comparison to YARRRML [
Once the RML mapping rule is generated, a platform-specific KGCP is established. This KGCP can
be reused to automatically generate KGs from various datasets and across different platform instances.
While the JSON preprocessor ensures the general applicability of the KGCP by normalizing the data
from IoT platforms, the RML engine, MorphKGC [
The proposed framework is planned to be applied in the building sector as illustrated in Figure 2. Currently, the deployment of promising smart building applications—such as for automation, fault detection, and energy monitoring—is often a laborious task due to the heterogeneity of underlying building systems and their IoT platforms. Semantic-IoT addresses this challenge by facilitating the generation of KGs, which provide a unified, semantic representation of the available information. By leveraging SWT, such as reasoning and SPARQL, the deployment process can ultimately be automated.
The proposed framework is demonstrated through a building automation use case5. In this use case, an IoT system has been developed for hotel buildings based on the FIWARE platform. Platform-specific data 5https://github.com/N5GEH/semantic-iot/tree/main/examples/fiware
Semantic-IoT Framework
models6 have been designed in accordance with the NGSIv2 specification of FIWARE. To fully unlock the flexibility to connect different building automation services, the Semantic-IoT framework is used.
With the RML Generation module, twelve resource types are identified from the data models, including locations, sensors, and actuators. Brick7, a widely-used ontology for building energy systems, is used to provide the semantic foundation for the terminology matching. Although Brick can theoretically be applied to all identified resource types, the current terminology matching achieves less than 60% accuracy. For example, while the resource type PresenceSensor should ideally match the class brick:Occupancy_Count_Sensor, the similarity score between them is only 0.5. Consequently, identifying suitable terminologies remains largely a manual work. In total, the complete RML mapping rules consists of 344 lines, with manual intervention limited to 18 lines for validation and 7 lines for manual input. Thus, the proposed framework significantly simplifies the declaration of RML mapping rules for IoT platforms.
With the generated RML mapping rules, a KGCP for the FIWARE-based platform is established, enabling the construction of KGs for any hotel that utilizes the same platform. To test its applicability, we then provisioned a range of virtual hotels, from a small 2-room layout to a large 1000-room one. For each hotel, we fetched the available data from the FIWARE NGSIv2 API to create the test datasets8. As a result, the KGCP generates KGs that represent the hotel buildings—including their rooms, sensors, and actuators—and integrate URLs for data interaction via the platform API.
Subsequently, we demonstrate the possibility to automate the deployment of automation services for ventilation control9. The tasks are mainly twofold: first, to decide on control strategies based on the available sensors and actuators; and second, to establish reliable data interactions via the platform API. To optimize information retrieval, we employ OWL-RL[9] to infer the generated KGs. Numerous class subsumption is added; for example, the class brick:Ventilation_Air_System is inferred to be a subclass of brick:Air_System and brick:HVAC_System. Thus, generalized SPARQL queries can be applied to retrieve any possible actuators in the hotel air systems. The availability of sensors can also be queried in a similar way. As a result, configurations for the ventilation controller can be automatically generated. 6https://github.com/N5GEH/n5geh.data_models/tree/main/example_building_automation 7https://brickschema.org/ 8https://github.com/N5GEH/semantic-iot/tree/main/examples/fiware/hotel_dataset 9https://github.com/N5GEH/semantic-iot/tree/main/examples/fiware/application_deployment
In this paper, we introduce a framework that automates the generation of knowledge graphs (KGs) from IoT platforms. By integrating system information and data interactions into the generated KGs, our approach enhances interoperability across diverse IoT systems, thus simplifying the development and deployment of cross-platform applications.
In future work, we aim to leverage the HTTP Vocabulary10 to enrich the semantic representation of platform APIs. Moreover, the current string similarity based terminology matching exhibits limited accuracy. Consequently, we plan to investigate alternative approaches, such as word embedding models. Ultimately, we intend to conduct comparative case studies that deploy advanced building automation programs across different IoT systems with various representative IoT platforms.
We gratefully acknowledge the financial support provided by the Federal Ministry for Economic Affairs and Climate Action (BMWK), promotional reference 03EN1030B.
During the preparation of this work, the authors used Gemini in order to: Grammar and spelling check, Paraphrase and reword. After using these tools/services, the authors reviewed and edited the content as needed and take full responsibility for the publication’s content. [8] D. Calvanese, G. D. Giacomo, D. Lembo, M. Lenzerini, R. Rosati, Ontology-Based Data Access and Integration, in: Encyclopedia of Database Systems, Springer, New York, NY, 2018, pp. 2590–2596. doi:10.1007/978-1-4614-8265-9_80667. [9] I. Herman, OWL-RL: OWL-RL: A simple OWL2 RL reasoner on top of rdflib, Zenodo, 2014. URL: https://doi.org/10.5281/zenodo.14543. doi:10.5281/zenodo.14543.