Semantic interoperability is an issue in heterogeneous IoT systems. The limited processing power and memory storage of constrained IoT nodes prevents them from handling enriched data. This paper proposes a method to lower complex knowledge representations into simpler structured data, based on the reuse of lifting mappings from data schemas to semantic models.
This method is based on the reuse of existing mappings used for generating enriched data instead of dedicated ones, thereby reducing the cost of lowering. A qualitative evaluation of our approach is then provided. 2
Enrichment and lowering: data management in the IoT 5 http://onem2m.org/ 6 https://allseenalliance.org/ enrichment meets a set of requirements, the mappings can be used for knowledge lowering as well, cutting by half the work required to make constrained devices semantically enabled. 3 A pivot-tree based approach to mapping reversal-based knowledge lowering Our approach to schema alignment reversing is divided into two steps: transformation of an RDF graph into an abstract pivot tree, and serialization of this pivot tree into a concrete syntax. The pivot tree is a language-agnostic representation of the nal tree (either XML, JSON or an URI for REST interfaces) with no concrete syntax. The pivot tree associates the knowledge expressed in the RDF graph (with its explicit semantics) to the tree-like structure of the targeted schema, so that its semantics is no longer implicit.Algorithm 1 shows the creation process of the abstract tree from the annotated schema and the target ontology.
Algorithm 1 Abstract tree creation 1: procedure buildAbstractTree(Schema $s, Graph $g, Resource $individual) match is a function that nds the schema element corresponding to the root individual 2: $schemaRoot match($s; $individual) 3: build constructs an abstract node from an rdf node and a matching schema node 4: $abstractRoot build($schemaRoot; $individual) 5: for $property in $graph.triples($individual, $property, $object) do 6: if $schemaRoot.hasAnnotatedDescendant ($property) then 7: if $property is ObjectProperty then 8: $abstractN ode buildAbstractTree($schemaRoot; $g; $object) 9: $abstractRoot:buildObjP rop($s; $g; $property; $abstractN ode)
This approach makes assumptions: the knowledge and the schema must meet some requirements. The target format must be a tree-like structure aligned to semantic resources, schema root(s) must be mapped to classes and meaningful inclusions in the schema must be mapped to relationships. There should be no ambiguity in the mapping: when transforming the RDF graph into the abstract tree, every relation from a node in the graph should be associated to one relation from the corresponding node in the schema.
To evaluate our solution, we measured the compliance of existing schemas with our requirements, and conducted a qualitative comparison with existing works (cf. g. 1). Di erent sources of data schemes were used to assess the hypothesis for a reversable mapping : schemes provided as part of standards speci cations (AllJoyn), and a sample of schemes from an online API repository9. Overall, 9 http://www.programmableweb.com these schemes met our requirements, or needed minimum backward-compatible transformation (such as the semantic annotation for enrichment). Supported Code Execution Generation Supports Flexible Schema type Mapping scenario base time required inference applicability required tDrairnescftorsmchaetmioan Transformation Small Fast No No No Any to any None
Schema
transformation Transformation Large Fast Yes No Partially Annotated schema Two-way
generation [
To bring semantic interoperability to constrained devices, data enrichment and knowledge lowering are required. We propose in this paper a knowledge lowering technique that reuses schema mappings dedicated to data enrichment in order to reduce the cost of integrating constrained devices in an IoT network. Future work will include a more complete evaluation, as well as an integration of this approach to an autonomic agent for the control of a connected apartment.