FUnctionality Sharing In Open eNvironments
Heinz Nixdorf Chair for Distributed Information Systems

Scrutable Adaptivity in Community-Enabled Web Portals

Title: Scrutable Adaptivity in Community-Enabled Web Portals
Authors: Fedor Bakalov
Source: PhD Thesis
Place: University of Jena, Germany
Date: 2012-09-01
Type: Dissertation

Personalized systems emerged as a solution to the problem of steadily
growing amounts of information and constantly increasing complexity of
navigation in the information space that overwhelms users. These systems
are able to learn about the needs of individual users and to tailor the
content, appearance, and behavior to the user needs. Web portals are
one of the earliest adopters of personalization technology. Examples of
personalized portals range from news aggregating portals that recommend
interesting news stories identified based on the stories that the user
read in the past to e-commerce portals recommending products identified
based on the user’s previous purchases. The aim of such adaptive
behavior is to help users to find relevant content easier and faster.

achieve this type of adaptivity, an adaptive system needs a number of
personalization models. More precisely, it needs a user model providing
information about users, e.g., about their interests, expertise,
background, or traits. It also needs metadata of information resources.
To achieve automatic selection of the resources that match the user’s
individual needs, both the user model and metadata should use the same
vocabulary, which requires a domain knowledge model represented in a
formalism that can be interpreted by the portal. Finally, the system
needs some logic or rules that govern how the resources must be
delivered given the user model and metadata.

In most
personalized systems, these models and the process of adaptation are
hidden from users. Users see only the resulting personalization effects.
But they do not have direct access to the information the system
collects about them and do not have control over the personalization
behavior. This results in a number of grave usability and privacy
problems. It violates such principles as predictability, transparency,
controllability, and unobtrusiveness. Also, it violates privacy
legislation and negatively impacts the trustworthiness of adaptive
systems. Finally, preventing users from controlling personalization may
cause that the models that the system uses for adaptation are incomplete
or contain obsolete or inaccurate information. This, in its turn, leads
to wrong adaptive behavior and irrelevant recommendations.

this thesis, we propose an approach to solve the above-mentioned
problems. We present a framework for scrutable adaptivity in
community-enabled web portals.  This framework provides four models that
an adaptive system needs for personalization: (1) a user model
representing information about individual users, (2) a metadata
repository providing annotations of content, (3) domain model defining
machine-processable semantics of the domain knowledge used for modeling
user features and annotating content, and (4) personalization rules
defining the logic of adaptation. Also, the framework provides methods
for keeping these models up-to-date, complete, and accurate. Among
others, these methods leverage the technology for Natural Language
Processing and the willingness of the user community to contribute and
annotate content. Furthermore, the framework provides graphical user
interfaces and interaction patters for allowing users to view and adapt
these models.

One of the key contributions of this research is a
novel interface for scrutinizing large and complex semantic user models
IntrospectiveViews. The interface visualizes such models leveraging a
metaphor of circular zones partitioned into slices, where each zone
represents items of certain interest degree and each slice represents
items of a specific type. The visualization provides functions for
getting overview, zooming, filtering, navigation, and search. It also
displays relevant content and semantic relations among items. In
addition to viewing, IntrospectiveViews allows users to edit the model
in an easy-to-use and efficient manner. It allows adding and deleting
items, changing status of items, organizing items by type, defining user
own types, and creating semantic relations among items. In this thesis,
we report on several users studies of IntrospectiveViews. Results of
these studies show that users deem the visualization
simple-to-understand, user friendly, visually attractive, and engaging.

important contribution of our research is a visual method for
scrutinizing personalization rules and effects in web portals. Our
approach gives users an overview of all adaptations that the portal
makes. Also, it allows users to influence these adaptations by either
changing personalization rules or editing the user model. The tool for
scrutinizing personalization is placed in the direct proximity to the
personalized content. Any change that the user makes on personalization
rules or the user model takes effect immediately on the personalized
content. This helps users to understand the connection between these two
components and the end personalization effects. Also, it allows them to
easily adjust the adaptive behavior to their needs and preferences. Our
approach can be used for fine-tuning of personalization effects at the
portlet level. Users can customize personalization rules and, if
necessary, deactivate personalization completely for each portlet

Finally, we report on a user study evaluating
impacts of our approach as a whole on the user’s perception of
personalized portals. In this study, we compared a personalized portal
without support for scrutable adaptivity with an identical personalized
portal in which such support was provided following our approach. 
Results of this study show that our approach makes significant impacts
on a number of aspects of personalized portals. It makes the adaptivity
more transparent and understandable. It improves the usefulness of
personalized portals and makes users be more satisfied with them.

File: dissertation.pdf