SEMANTIC WEB

On the web there is need to equip content with some meta-information about its
meaning (semantics). Mostly used semantic (and also knowledge) representation are
ontologies. Tagging information with its meaning is useful for many reasons: content is
understandable and reusable by machines [6] and can be easily shared among computer
systems. Another profit of semantics is in information searching, because it enables
searching by meaning not only by keywords contained in content.
Further ontology topics are knowledge reasoning for deriving new facts from
ontologies and knowledge processing used e.g. for extracting semantics from unstructured
text. Another inspirable term from human brain researches is cognitive flexibility defined as
ability to restructure knowledge spontaneously according to adaptive response.
There is a challenge how to appropriately represent various types of knowledge and
domains and how to interact these types when generating adaptive content. Everyone can
express same thing differently.



ADAPTIVE WEB
Adaptive web is advanced content providing and presenting technique. It provides
information on the web according to specific needs of users or user groups [2]. Adaptive
web is opposite concept than unified generic content, same for everyone. There are two
crucial problems in adaptation, at first how to identify these needs and second how to map
those needs to available resources. Most important usage of adaptation is in e-learning.
Learning courses can be adapted to each student's preferences, foreknowledge, learning
progress and cognitive style. Let's give some examples: In first [3] we have ontology,
which is branching structure, representing some knowledge. Then we can generate
learning content, such as slides, in reasonable order using preorder DFS of ontology class'
tree and we can also generate multi-choice test with class-subclass and class-instance
questions derived from ontology. In second example adaptive web can provide different
amount and detail of content to nurse and to medicine doctor according to their
foreknowledge.

Adaptive web technology is also applicable in recommender systems, which analyze
usage of entities (services, products etc.) in order to point out additional relevant and
interesting information or entities. In sales terminology this is called cross-selling.
There are three levels of adaptation:
presentation
content
navigation
In presentation level type of media (e.g. text, audio) or its form (e.g. visualization,
structure) is adapted. Blind user receives audio content instead of text; graphical symbols
are replaced by text description for text terminals etc.
In content level amount and detail of information is adapted. Alternatively completely
different content can be provided for different inputs. Example of content adaptation was
mentioned above.
Last level of adaptation is navigation one. Links are enabled or disabled, visible or
hidden, reordered or annotated according to appropriate or recommended next steps in
content absorbing by user. When system realizes that all prerequisites are learned,
following course can be recommended. Well-known type of navigation adaptation is
semaphore technique. Links are annotated by red, yellow or green symbol according to
appropriateness of such link for current user.




Information is adapted by user and context inputs. User input is related to user model
ontology mentioned below. Main factors of user inputs are his/her preferences, desired
goals, cognitive and learning style, foreknowledge and learning progress.
Context input is related to context model ontology. It can be very close to user input
in some cases. Let's give some examples: content can be adapted to device and its
properties e.g. PDA or mobile phone due to its limited-size display, in second example
presentation can be adapted according to user handicap e.g. audio representation for blind
user instead of text. We can discuss whether handicap is context or user input, but we
consider division of user as inner input and context as outer input and reason this way that
handicap is not something that comes from user inside.
There are many models for adaptive content (e-learning) systems such as SACS
model [1] or AHAM.
In model according to Simple Adaptive Content System (SACS) there are two user
roles: Administrator and Student. Administrator configures the system and adds new
content. Student is browsing available content and his/her study state is monitored by
system in order to offering most appropriate content,
WHAT is adapted
ACCORDING TO WHAT
HOW is adaptation made
In earlier model WHAT was represented by domain model, describing adaptable
content, ACCORDING TO WHAT was represented by user model, describing user
preferences, foreknowledge, cognitive and learning style and learning progress. Finally HOW was represented by adaptation model, rules that defines adaptation process itself. In later model according to new needs model was modified: ACCORDING TO WHAT is divided into user and context model and HOW is divided into activity and adaptation model (see Figure 2). Context model describes outer inputs entering to adaptation
process. There can be privacy issue in adaptive web, because of conflict between most possible adaptation of content according to individual user on the one side and his/her privacy on the other side. User can get feeling that systems know everything about him/her and knows all his/her secrets and pains. This problem was discussed about adaptable search engines such as Google.

ONTOLOGY
Sampson [6] defines ontology as representation of shared conceptualization of
particular domain and declares it as major semantic web component.
Because of instances are not reusable, we need to represent knowledge itself [5].
Usual representation of knowledge is branching structure e.g. concept graph, semantic net or mind-map, so ontology is also branching, defining classes and properties. In graph representation generating of new domains can be made using graph transformations. There are many ontology models used, usually divided into declarative and procedural knowledge. Declarative knowledge includes domain model (content, WHAT level of AHAM) and user and context models (ACCORDING TO WHAT). Procedural knowledge includes adaptation and activity models (HOW).
User model captures learner profile, and provides the information on which adaptivity
is based [3]. There are many specifications to describe user model based on XML e.g.
UserML, which was not accepted by community anyway, OWL and RDF(S) and some special like PAPI, LIP and ACCLIP [5]. User model can be predefined, e.g. according to expected profiles of user or adaptive according to individual user’s behavior. User model can learn from actively provided information (like configuring preferences or collaborative tagging) or passively from user’s interaction and behavior.
Domain model is internally represented by RDF(S) or special Web Ontology
Language (OWL). Domain model (and even user model) can be also imported from relation database [7]. It could be hard to describe domains appropriately because of its vast heterogeneity. OWL is used in system learning techniques, where system is
generating new ontology or is validating existing one. There are some tools for designing ontology e.g. Protégé.
Context model interfere with user model, we can consider user's handicap is in
context model instead of user model. Another example of context model is target device dependence (content needs to be adapted for display with reduced color depth or for device with limited transfer rates). Activity and adaptation model are procedural types of knowledge and both describe HOW adapt the content. Adaptation model is set of rules usually mapping domain, user
and context model (domain) to adapted content parts (range). For its description rule
based languages such as Lisp or Prolog is used or there are special semantic web
languages e.g. RIF (Rule Interchange Format), which is used mainly for interchange and rule languages conversion; SWRL (Semantic Web Rule Language) or OCL (Object Constraint Language).
We should mention importance of ontology mapping and merging [4]. Because of
various types of domains and their using, everyone can focus different detail of object and create different ontology. Mapping helps to find relevant classes in both ontologies. In this way mapping and merging enables better knowledge sharing. There are some tools for ontology mapping e.g. Pellet plug-in into Protégé ontology designing software.