AN APPROACH TO MORE RELIABLE CONTEXT-AWARE
SYSTEMS BY ASSESSING AMBIGUITY
Taking into Account Indetermination and Vagueness in Smart Environments
Aitor Almeida and Diego López-de-Ipiña
Deusto Institute of Technology - DeustoTech, Universidad de Deusto, Bilbao, Spain
Keywords: Fuzzy Logic, Uncertainty, Vagueness, Context, Data Fusion, Inference.
Abstract: Often context-aware systems consider the environment a defined element. Meanwhile reality is full of
vagueness and uncertainty. Taking into account these aspects we can provide a more grounded and precise
picture of the environment, creating context-aware systems that are more flexible and reliable. It also
provides a more accurate inference process, making possible to consider the quality of the context data. In
order to tackle this problem we have created an ontology that considers the ambiguity in smart
environments and a data fusion and inference process that takes advantage of that extra information to
provide better results.
1 INTRODUCTION
Intelligent environments host a diverse and dynamic
ecosystem of devices, sensors, actuators and users.
Modelling real environments taking certainty for
granted is usually a luxury that a context
management framework cannot afford. Reality, and
hence the context, is ambiguous. Sensors and
devices are not perfect and their measures carry a
degree of uncertainty, several thermometers in the
same room can provide conflicting measures of the
temperature and there always exists the human
factor. Not every user can provide the exact
temperature they want for their bath, most of them
will only say that they want it “warm”. For this
reason, when developing smart spaces and ambient
intelligence application, it is important to address
ambiguity in order to model more realistically the
context. To provide our systems with this feature,
we have centred our work in two aspects of the
ambiguity: uncertainty and vagueness. We use
uncertainty to model the truthfulness of the different
context data by assigning to them a certainty factor
(CF). This way we can know the reliability of each
piece of information and act accordingly. These data
also allow us to create a more robust data fusion
process to resolve the problem of the existence of
multiple providers for the same piece of information
in the same location. On the other hand, vagueness
helps us to model those situations where the
boundaries between categories are not clearly
defined. This usually occurs when users are
involved. Different users will have different
perceptions about what is a cold room or a noisy
environment. We have addressed this problem using
fuzzy sets to model the vagueness.
In this paper we will describe the three main
components of the ambiguity conscious frameworks
we have developed. First we will describe the
ontology created to model the uncertainty and
vagueness in context. Then we will discuss the data
fusion process that takes place to infer the real status
of the rooms using multiple measures. Finally we
will describe the implemented inference mechanism
that processes ambiguity as a whole, combining
vagueness and uncertainty.
2 RELATED WORK
Several authors have worked into combining
indetermination or vagueness with ontologies. An
extensive survey can be found in (Lukasiewicz and
Straccia, 2008). In the case of the indetermination, in
(da Costa et al., 2005) authors present a probabilistic
generalization of OWL called PR-OWL based in
MEBNs (Multi Entity Bayesian Networks) which
allows the combination of first order logic with
233
Almeida A. and López-de-Ipiña D..
AN APPROACH TO MORE RELIABLE CONTEXT-AWARE SYSTEMS BY ASSESSING AMBIGUITY - Taking into Account Indetermination and
Vagueness in Smart Environments.
DOI: 10.5220/0003800502330236
In Proceedings of the 2nd International Conference on Pervasive Embedded Computing and Communication Systems (PECCS-2012), pages 233-236
ISBN: 978-989-8565-00-6
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
Bayesian logic. This ontology represents the
knowledge as parameterized fragments of Bayesian
networks. In (Ding et al., 2006) authors propose
another probabilistic generalization of OWL called
BayesOWL which also uses Bayesian networks.
Authors suggest a mechanism which can translate an
OWL ontology to a Bayesian network, adding
probabilistic restrictions when building the network.
The created Bayesian network maintains the
semantic information of the original ontology and
allows ontological reasoning modeled as Bayesian
inference. (Yang and Calmet, 2005) describe another
integration of OWL with Bayesian networks, a
system named OntoBayes. It uses an OWL
extension annotated with probabilities and
dependencies to represent the uncertainty of
Bayesian networks. Several authors have also
addressed the combination of the vagueness
(represented as the usage of fuzzy sets) with
ontologies. In (Stoilos et al, 2005) authors analyze
how SHOIN could be extended adding the
possibility of using fuzzy sets (f-SHOIN). They also
propose a fuzzy extension for OWL. In (Bobillo and
Straccia, 2009) authors describe a fuzzy extension
for SROIQ(D) and present an Fuzzy OWL2
Ontology. In (Parry, 2004) a fuzzy ontology for the
management of medical documents is discussed.
This ontology can store different membership
values. Additionally the author has created a
mechanism based in the occurrence of keywords in
the title, abstract or body of the document to
calculate the membership value of the different
categories. In (Lee et al., 2005) authors describe a
fuzzy ontology used to automatically create
summaries of news articles. Authors have also
created a mechanism for the automatic creating of
the fuzzy ontology based on the analysis of the
news.
The work discussed in this paper combines both
approaches to model the ambiguity
3 AMBI
2
ONT ONTOLOGY
One of the problems we encountered modelling
context data in previous projects was the use of the
uncertainty and vagueness of the gathered
information. In the Smartlab project (Almeida et al.,
2009) none of this information was used, which led
to a loss of important data like the certainty of the
measures taken by the sensors. In the Imhotep
framework (Almeida et al., 2011) we started using
fuzzy terms to describe a small part of the context
(the capabilities of mobile devices and users) in a
human-friendly manner. Our objective with the work
presented in this paper was to develop a framework
capable of managing the ambiguity and incertitude
that often characterizes the reality. To do this we
have created an ontology that models these concepts.
The main elements of the ontology are: 1) Location:
The subclasses of this class represent the location
concepts of the context. 2) LocableThing: The
subclasses of this class represent the elements of the
system that have a physical location. 3)
LinguisticTerm: This class models the fuzzy
linguistic terms of the values of the context data.
The ontology only stores the linguistic term and
membership value of each individual of context
data. 4) Capability: The subclasses of this class
model the capabilities of users and their mobile
devices. One objective of our framework is to be
integrated with the Imhotep Framework that allows
creating adaptive user interfaces that react to these
capabilities and the changes on the context. Our
ontology models two aspects of the ambiguity of the
context data, the uncertainty (represented by a
certainty factor, CF) and the vagueness (represented
by fuzzy sets). Uncertainty models the likeliness of a
fact, for example “the temperature of the room is
27ºC with a certainty factor of 0.2 and 18ºC with a
certainty factor of 0.8” means that the value of the
temperature is more probably 18ºC (but it cannot be
both of them). In the case of vagueness it represents
the degree of membership to a fuzzy set. For
example “the temperature of the room is cold with a
membership of 0.7” means that the room is mostly
cold. Each ContextData individual has the following
properties: 1) Crisp_value: the measure taken by the
associated sensor. In our system a sensor is defined
as anything that provides context information. 2)
Certainty_factor: the degree of credibility of the
measure. This metric is given by the sensor that
takes the measure and takes values between 0 and 1.
3) Linguistic_term: each measure has its fuzzy
representation, represented as the linguistic term
name and the membership degree for that term.
4 AMBIGUOUS SEMANTIC
CONTEXT
The semantic context management is done in four
steps: 1) add the measures to the ontology, 2)
process the semantic and positional information, 3)
apply the data fusion mechanism and 4) process the
ambiguity contained in the data.
To add a measure to the ontology the sensor
PECCS 2012 - International Conference on Pervasive and Embedded Computing and Communication Systems
234
must provide the measure type, its value, location
and a certainty factor. We assume that each sensor
knows its certainty factor based on its type and
manufacturer. We also assume that the certainty
factor of the sensor can change over time depending
on the environment (e.g. a thermometer can be
pretty accurate for temperatures between -10ºC and
50ºC but the measure quality can degrade outside
that range). For that reason the sensor certainty
factor is not stored in the ontology when the sensor
registers itself, it is provided with each measure.
Once the measures have been added, we apply a
semantic inference process to achieve two goals:
make explicit the hidden implicit knowledge in the
ontology and infer the positional information of each
measure. To do this we use two different sets of
rules: the semantic rules and the spatial heuristic
rules. To make the semantic reasoning less
cumbersome we implement a subset of the RDF
Model Theory and the OWL Model. The spatial
heuristic rules are used to infer higher level
positional information from the coordinates provided
by the sensors. This information comprises data like
the room in which the sensor is located; the devices,
people and sensors surrounding it and the relative
location to other LocableThing-s (refer to section 3
for more information about the elements of the
ontology).
Once the location and semantic information of
the measures has been inferred and processed the
data fusion process is applied. From the previous
step we can infer that each room can have multiple
sensors that provide the same context data (e.g
various thermometers in the same room). Usually
the values and certainty factor of those measures do
not coincide. To be able to take the proper actions
we need to process those differing measures to
assess the real status of to room. To tackle this
problem we have created a data fusion mechanism
that refines those individual measures into a single
global measure for each room. We have
implemented two types of strategies for this process:
tourney and combination. Using the tourney strategy
the measure with the best CF is selected as the
global measure of the room. On the other hand the
combination strategy has three different behaviours
as stated in (Bloch, 1996): 1) Severe, the worst
certainty factor from all the input measures is
assigned to the combined measure; 2)Indulgent, the
best certainty factor from all the input measures is
assigned to the combined measure; 3)Cautious, an
average certainty factor is calculated using the
certainty factor from the input measures.
To determine the combined measure value we
weight the individual values using their certainty
factors as seen in the following equation.
=
∑(
∗
)
∑
(1)
Where m is the measure values and cf is the measure
certainty factor. Finally we process the ambiguity.
As explained previously we model two aspects of
the ambiguity: the uncertainty and the vagueness. To
be able to reason over this information we have
modified the JFuzzyLogic Open Source fuzzy
reasoner to accept also uncertainty information.
JFuzzyLogic follows the FCL standard for its rule
language. The modified reasoner supports two types
of uncertainty, uncertain data and uncertain rules.
The first type occurs when the input data is not
completely reliable (as seen in the example shown in
Table 1). To support this type of uncertain data we
have modified the API of the reasoner. The second
type of uncertainty takes place when the outcome of
a rule is not fixed, for example “if the barometric
pressure is high and the temperature is low there is a
60% chance of rain”. To model this aspect of
uncertainty we have modified the grammar of the
FCL language. Uncertainty and fuzziness can appear
in the same rule and influence each other. To tackle
this problem we have implemented the inference
model described in (Orchard, 1998). This model
contemplates three different situations depending on
the nature of the antecedent and consequent of the
rule and the matching fact: CRISP Simple Rule
where both antecedent and matching fact are crisp
values, FUZZY_CRISP Simple Rule where both the
antecedent and matching fact are fuzzy and the
consequent is crisp and finally the FUZZY_FUZZY
Simple rule where all three are fuzzy. In the case of
the CRISP Simple Rule the certainty factor of the
consequent is calculated using the following
formula:
=
×
(2)
Where CFc is the certainty factor of the consequent,
CFr is the certainty factor of the rule and CFf is the
certainty factor of the fact. In the case of
FUZZY_CRISP Simple Rule the certainty factor of
the consequent is calculated using the following
formula:
=
×
×
(3)
Where S is the measure of similarity between both
fuzzy sets and is calculated using the following
formula:
=
(
|
)
(
|
)
AN APPROACH TO MORE RELIABLE CONTEXT-AWARE SYSTEMS BY ASSESSING AMBIGUITY - Taking into
Account Indetermination and Vagueness in Smart Environments
235
=(
(
|
)
−0.5)×
(
|
)
ℎ
(4)
Where:
(
|
)
= maxminμ
(u),μ
(u), ∀ ∈
(5)
And:
(
|
)
=1−
(
|
)
(6)
Finally in the case of FUZZY_FUZZY Simple Rule
the certainty factor of the consequent is calculated
using the same formula than in the CRISP Simple
RULE. Currently we do not support this type of
combined reasoning for complex rules that involve
multiple clauses in their antecedent.
5 CONCLUSIONS AND FUTURE
WORK
We have presented in this paper a context-aware
system that takes into account the uncertainty and
vagueness present in smart environments. We have
also described an ontology to model this ambiguity.
The presented system provides a more detailed
picture of the environment, allowing a richer
reasoning over the context. We have also described a
data fusion mechanism applied in the case that
multiple data sources for the same measure exist in
one room. This mechanism relies on the uncertainty
information provided by our system to create a
global assessment for each room that tries to infer
the real situation. Our final goal is to provide a more
robust and flexible mechanism to manage the
context, that allows capturing richer nuances of the
environment.
As future work, first we would like to create a
mechanism that automatically assesses the certainty
factor of a sensor comparing its data with the one
provided by other sensors. This will allow us to
identify and discard malfunctioning sensors
automatically. Secondly we would like to develop an
ecosystem of reasoners to distribute the inference
process. We hope that this distribution will lead to a
more agile and fast reasoning over the context data,
allowing us to combine less powerful devices to
obtain a rich and expressive inference.
ACKNOWLEDGEMENTS
This work has been supported by project grant
TIN2010-20510-C04-03 (TALIS+ENGINE), funded
by the Spanish Ministerio de Ciencia e Innovación.
REFERENCES
Lukasiewicz T., Straccia U. Managing uncertainty and
vagueness in Description Logics for the Semantic
Web. Journal of Web Semantics, 6(4), 291-308. 2008.
P. C. G. da Costa, K. B. Laskey, and K. J. Laskey. PR-
OWL: A Bayesian ontology language for the Semantic
Web. In Proceedings URSW-2005, pages 23–33, 2005.
Z. Ding, Y. Peng, and R. Pan. BayesOWL: Uncertainty
modeling in Semantic Web ontologies. In Z. Ma,
editor, Soft Computing in Ontologies and Semantic
Web, volume 204 of Studies in Fuzziness and Soft
Computing. Springer, 2006.
Y. Yang and J. Calmet. OntoBayes: An ontology-driven
uncertainty model. In Proceedings IAWTIC-2005,
pages 457–463. IEEE Press, 2005.
Stoilos, G. and Stamou, G. and Tzouvaras, V. and Pan,
J.Z. and Horrocks, I. Fuzzy OWL: Uncertainty and the
semantic web. Proc. of the International Workshop on
OWL: Experiences and Directions. Vol 280, 2005.
Bobillo, F. and Straccia, U. An OWL ontology for fuzzy
OWL 2. Foundations of Intelligent Systems, 151—
160. 2009.
Parry, D. A fuzzy ontology for medical document
retrieval. Proceedings of the second workshop on
Australasian information security, Data Mining and
Web Intelligence, and Software Internationalisation-
Volume 32, pp 121-126. 2004.
Lee, C. S. and Jian, Z. W. and Huang, L. K. A fuzzy
ontology and its application to news summarization.
Systems, Man, and Cybernetics, Part B: Cybernetics,
IEEE Transactions on. Vol 35, n5, pp859—880. 2005.
Almeida A., Orduña P., Castillejo E. et al. Imhotep: an
approach to user and device conscious mobile
applications. Personal and Ubiquitous Computing
(Journal), Volume 15, Number 4, pages 419-429.
Springer. April 2011.
Almeida A., López-de-Ipiña, D., Aguilera U. et al. An
Approach to Dynamic Knowledge Extension and
Semantic Reasoning in Highly-Mutable Environments.
3rd Symposium of Ubiquitous Computing and Ambient
Intelligence, Advances in Soft Computing, vol 51,
pp265-273, 2009.
Bloch, I. Systems, Man and Cybernetics, Part A: Systems
and Humans, IEEE Transactions on, vol 26, num 1, pp
52-67. 1996.
R. A. Orchard. FuzzyCLIPS Version 6.04A User’s Guide.
Integrated Reasoning Institute for Information
Technology National Research Council Canada. 1998.
PECCS 2012 - International Conference on Pervasive and Embedded Computing and Communication Systems
236
