Introduction
more advanced semantic matchmaking [
2
] w.r.t. other reasoning engines.
MiniME uses the OWL API [
5
] to parse and manipulate Knowledge Bases in OWL 2
supported syntaxes. It exploits structural inference algorithms on unfolded and
CNF (Conjunctive Normal Form) normalized concept expressions for e cient
computations also on resource-constrained platforms. In [
17
] four Semantic Web
reasoners were successfully ported to the Android platform, albeit with
significant rewriting or restructuring e ort in some cases. Similarly, in [
6
] the ELK
reasoner was optimized and evaluated on Android. Nevertheless, all those
systems were designed mainly for batch jobs over large ontologies and/or expressive
languages, which made mobile devices less suitable due to slower computation
and smaller memory. The non-standard services of Mini-ME are more useful
in SWoT scenarios, where mobile agents provide quick decision support and/or
on-the- y organization in environments intrinsically unpredictable as the mobile
ones. Mini-ME has been now updated, including novel features (see Section 2
for details). Main improvements comprise: (i) optimization for a more e cient
memory management; (ii) software re-engineering for improved maintainability;
(iii) support for the OWLlink protocol [
7
]; (iv) introduction of abduction-based
Concept Covering inference service, described in Section 2.1; (v) implementation
of a pair of plug-ins for the Protege ontology editor, described in Section 2.2.
Mini-ME was employed in several prototypical testbeds in the eld of
Semantic Web of Things. Two use cases are presented here as proof of concept:
Section 3.1 reports on a wireless semantic sensor network based on CoAP
(Constrained Application Protocol) [
1
], while Section 3.2 overviews a mobile semantic
augmented reality scenario. Section 4 closes the work.
2
Improvements and novel features
In its early version Mini-ME exposed two interfaces: OWLReasoner and
MicroReasoner [
12
], respectively for standard and non-standard inference services.
Now they have been consolidated in the latter interface which provides all the
services entry points. In addition, data structures have been organized in a
hierarchy which allows more exible manipulation and e cient memory usage.
Support for the OWLlink [
7
] protocol was integrated, based on the OWLlink
API [
9
]. OWLlink allows OWL 2 [
16
] reasoners to o er a standard
HTTP/XMLbased interface to applications. OWLlink support is currently limited to the core
protocol, thus allowing requests for standard inferences only. Extensions for
nonstandard inference services are being devised and are planned for integration in
the next Mini-ME version. Further novel features are described in the following
subsections and in the Mini-ME web page at http://sisin ab.poliba.it/swottools/
minime, where also the results of the OWL Reasoner Competition2 will be
published.
2 http://vsl2014.at/meetings/ORE-competition.html
Concept covering
Many SWoT scenarios require that relatively large number of low-complexity
resources are aggregated in order to satisfy an articulated request. To this aim,
in addition to Concept Abduction and Concept Contraction non-standard
inferences, a further reasoning task based on the solution of Concept Covering
Problem (CCoP, formally de ned in [
10
]) is now available. It allows to: (i) cover
(i.e., satisfy) features expressed in a request as much as possible, through the
conjunction of one or more instances of a Knowledge Base (KB) {seen as
elementary building blocks{ and (ii) provide explanation of the uncovered part of the
request itself. Given a concept expression R (request) and a set of instances S =
fS1, S2, ... , Sng (available resources), where R and S1, S2, ... , Sn are satis able
in the reference ontology T , Concept Covering aims to nd a pair hSc; Hi where
Sc includes concepts in S (partially) covering R w.r.t. T and H is the (possible)
part of R not covered by concepts in Sc. Algorithm 1 is applied to solve CCoP.
A compatibility check is performed (line 7) to verify if a resource Si (from set
S) can cover the request. Afterwards (line 8) abduce algorithm {described in
[
11
]{ solves a Concept Abduction Problem (CAP) to determine what is missing
in the resource description, in order to completely satisfy the request. It de nes
also a penalty value of Si w.r.t. H based on the norm of CNF expressions [
11
].
Finally, the resource (Smax) with the lowest penalty (rmin) {i.e., the resource
best covering H at each step{ is selected and moved from S to Sc (lines 17-18)
and the part of H covered by Smax is removed (line 19). The algorithm output
is the set of resources best covering the request, along with the uncovered part,
if present.
Algorithm 1 Algorithm for solving Concept Covering Problem (CCoP)
Algorithm: solveCCoP (hL; T ; R; Si)
Protege plugins
Mini-ME has been integrated within the Protege ontology editor [
4
] through the
implementation of an OWL reasoner plugin. It is accessible through the Protege
user interface in the Reasoning menu. A further Protege plugin has been
developed to exploit non-standard inferences through a user-friendly GUI (Figure 1).
It was devised to support users during the development of ontology for
pervasive scenarios; all supported inferences can be directly exploited and tested also
through Protege. The existing DL Query 3 plugin was used as guideline. The
proposed plugin is a Tab Widget and it consists of the following components,
highlighted in Figure 1: (A) OWLIndividualsList and OWLIndividualsTypes tabs,
showing all KB instances with related description; (B)
OWLAssertedClassHierarchy and OWLClassDescription tabs, containing the general taxonomy along
with the description of selected classes; (C) an input box used to select the
inference task to be executed, the request R and {in case of Concept Abduction and
Concept Contraction{ the resource annotation S. Both can be selected from the
OWLIndividualsList through drag-and-drop. For Concept Covering it is instead
possible to select a subset of KB individuals through the Individuals List panel
as composing resources; (D) (results area) shows the output of the selected
inference service. In Figure 1 a CCoP is solved, component individuals and the
uncovered part of the request are shown.
Motivating scenarios
CoAP-based semantic sensor networks
The Semantic Sensor Network (SSN) paradigm [
8
] aims at exploiting semantics
to increase exibility and interoperability in sensor networks. A novel SSN
framework was devised and proposed in [
14
], supporting resource discovery through
semantic matchmaking. It is based on: (i) a backward-compatible extension of
the HTTP-like Constrained Application Protocol (CoAP) [
1
] for resource
discovery; (ii) non-standard inference services for retrieving and ranking resources; (iii)
adoption of W3C standard SSN-XG ontology [
3
] to annotate data, events and
device features. Each sensor is basically seen as a server exposing both sensor
readings and internal information as resources toward clients, which act on
behalf of end-user applications. The standard CoAP resource discovery mechanism
only allows a syntactic string-matching of attributes, lacking explicit and formal
characterization of the resource semantics. A protocol enhancement has been
devised to support a logic-based matchmaking between a request and one or more
resource descriptions, both expressed using languages grounded on Description
Logics. In a car risk prevention scenario, semantic matchmaking was carried out
by running Mini-ME on a testbed comprising di erent Raspberry Pi embedded
boards with small computational capabilities, connected in a CoAP-based SSN.
Local or remote applications act as CoAP clients and use semantic-based
discovery to search for sensors or actuators, based on annotated descriptions of their
features. In standard CoAP a temperature sensor would be described just with
resource type rt=temperature and discovery would retrieve it only if request
exactly corresponded. On the contrary, in semantic-enhanced CoAP resource
type would be an OWL annotation w.r.t. a domain ontology and the request
semantics could be matched. Unfortunately, Subsumption test returns a yes/no
answer, so supporting only subsume (a.k.a. full) matches. Concept Abduction
and Contraction can also identify intersection-satis able (a.k.a. potential) and
disjoint (a.k.a. partial) matches, and also rank the resources according to the
degree of similarity w.r.t. request. In the experimental evaluation, Mini-ME showed
satisfactory performance in terms of processing time. Standard CoAP used on
average 150ms to reply a basic query for temperature sensors with two resources,
while 575ms were needed to perform a semantic CoAP discovery, executing
Concept Abduction on the same {semantically annotated{ resources and returning
results ranked by relevance w.r.t. the request.
3.2
Semantic-enhanced mobile augmented reality
Semantic-based technologies can support articulated and meaningful
descriptions of locations and Points of Interest (POIs). The use of metadata
(annotations) endowed with formal machine-understandable meaning can enable more
advanced location-based resource discovery through proper inferences. In a
previous work [
15
], a general method and a tool were presented for annotating
maps so allowing a collaborative crowd-sourced enrichment of OpenStreetMap
(OSM)4 basic cartography. In order to allow users to exploit enriched maps,
the framework is extended with a mobile Augmented Reality (AR) system for
semantic-enhanced POI discovery and exploration [
13
]. It allows users to see an
overlay of markers for POIs on the scene framed by their mobile device
camera. Exploiting the embedded Mini-ME matchmaker, the mobile tool executes
semantic matchmaking between the user pro le and the annotations of POIs
{embedded into semantic-enhanced OSM map{ in her surroundings, in a
reference range with respect to user's position. The user interface is shown in Figure
2a. It displays on a radar several semantic-enriched points of interest within a
radius which is adjustable through a slider on the right hand side. Matchmaking
outcomes are displayed as color-coded markers on the display used as device
camera view nder, corresponding to the real direction and distance of each POI
from the user. Markers for POIs within the eld of sight are also shown upon the
real-time device camera view. By touching a marker, the user can see its relevant
features, which are presented as icons around a wheel shape, in order to provide
a clear and concise description, as shown in the central portion (A) of Figure
2b. The View result panel (B) in Figure 2b lists all missing features w.r.t. user
pro le (C), computed through Concept Abduction. In case of incompatibility,
the same left-hand menu shows Concept Contraction outcome: properties the
POI satis es and incompatible elements (Figure 2c-(D)).
(a) User interface
(b) Abduction results
(c) Contraction results