Personalization Using Ontologies and Rules
Thanh Tran1, Haofen Wang2, Steffen Lamparter1, and Philipp Cimiano1
1 Institute AIFB, Universita¨t Karlsruhe, Germany
{dtr,sla,pci}@aifb.uni-karlsruhe.de
2 Department of Computer Science & Engineering
Shanghai Jiao Tong University, Shanghai, 200240, China
{whfcarter}@apex.sjtu.edu.cn
Abstract. Adaptive hypermedia systems can alleviate information overload on
the Web by personalising the delivery of resources to the user. These systems are
however afflicted with difficulties in the acquisition of user data as well as the
general lack of user control on and transparency of the systems’ adaptive behav-
ior. In this paper, we argue that the use of rules on top of ontologies can enable
adaptive functionality that is both transparent and controllable for users. To this
end, we sketch ODAS, a domain ontology for adaptive hypermedia systems, and
a model for the specification of adaptation rules.
1 Introduction
The vast amount of available information leads to confusion for the average user, mani-
fested by “comprehension and orientation problems” and a general “loss in information
space” [8]. Targeting at this problem, there are adaptable systems that allow users to
manually configure the resource provision. More advanced adaptive systems automati-
cally identify the information that is relevant to the user.
This adaptive behaviour is typically realised in commercial systems by collabora-
tive [4] and content-based filtering [5]. These filtering-based systems employ either a
user or a content model to recommend relevant information. The main drawbacks of
filtering-based approaches are well-known and have been discussed extensively in liter-
atures [9]. Content-based approaches lead to overspecialization, resulting in too much
recommendations of items of one specific type. In collaborative filtering, it is not possi-
ble to recommend a new item not yet rated by users. So, when there are few user ratings,
only a small set of items can be considered by the system for recommendation.
However, these drawbacks can be addressed when combining user with content re-
lated data [1]. In the same line, further data such as the user task, the environment and
the system have been incorporated. This extensive usage of contextual information as
well as the adoption of advanced machine learning techniques can improve adaptive
functionalities. However, apart from the inherent difficulties in collecting contextual in-
formation, adaptive systems are seen by the user as black-boxes, which give advices
but cannot be questioned. The underlying algorithms are built on latent factors and
heuristics that cannot be directly translated to explanations to facilitate the user in un-
derstanding the adaptive behavior.
In this paper, we propose ODAS as a domain ontology that improves the reuse and
exchange of context information (Section 2). ODAS allows to capture different aspects
W. Nejdl et al. (Eds.): AH 2008, LNCS 5149, pp. 349–352, 2008.
c
© Springer-Verlag Berlin Heidelberg 2008

350 T. Tran et al.
of the adaptation context such as the user, the task, the system, the environment, and
various aspects of the content—all contextual information that has been successfully
used to achieve sophisticated adaptive behaviour. In addition, we propose a model for
the specification of adaptation rules based on ODAS (Section 3). These rules capture
the logic of the adaptive behavior in a declaratively manner and hence, facilitate the
inspection and modification of the underlying adaptation model.
2 Towards a Domain Ontology for Adaptive Hypermedia Systems
In adaptive hypermedia systems, the use of contextual information of different types is
crucial to achieve effective adaptive behavior. We have identified different aspects of the
context, and incorporated into ODAS, an ontology we propose for the domain of adap-
tive hypermedia systems. It contains 138 subclass definitions, and a total of 504 axioms.
They have been specified using the standard Ontology Web Language (OWL) recom-
mended by the W3C. Figure 1 shows a portion of the ODAS concept hierarchy (black
arrows indicate that some concepts have been omitted). ODAS extends the well-known
top-level Suggested Upper Merged Ontology (SUMO). Also, ODAS has been aligned
with related domain ontologies and taxonomies, namely the Public and Private Informa-
tion (PAPI) [2], the IMS Learner Information Package (LIP), the Dublin Core metadata
scheme, its extension Learning Object Metadata scheme (LOM) [7] and the MPEG-7
ontology [3]. We consider this adoption of existing standards as crucial to achieve ac-
ceptance by the community and interoperability for the domain. We will now focus on
the main ODAS concepts that have been introduced to represent the context. They are
highlighted by rectangles in Fig. 1 and, henceforth, will be referred to as models.
Central to the representation of the adaptation context is the conceptApplication
Interaction. Basically, an instance of this concept establishes a context by connect-
ing different models. It tells the system that in an Application Environment
(environment model) a particular Cognitive Agent (user model) is currently in-
teracting with a Content embodied in a Content Bearing Object (resource
model) of the Application (system model) to accomplish a task (task model). A task
Computer-aided_Process
Content_Bearing_Object Social_Interaction
Agent
Mouse Keypad
Abstract_Entity
Touchpad
Concept Object
Artefact
Sentient_Agent
Physical_Entity
Network
Content
Entity
Proposition
Application_Interaction
Corpuscular_Object
Device
Voice
Application
Process
Region Intentional_ProcessSelf_Connceted_Object
Cognitive_Agent
Application_Environment Intentional_Psychological_Process
Fig. 1. ODAS Concept Hierarchy