Novel Approaches to Acquisition and Maintenance of User Model

Anton Andrejko

Doctoral dissertation project supervised by prof. Mária Bieliková


Motivation

There are many problems related to the Web — one aspect is that it is overloaded by huge amount of information where a searcher can get lost in information space, spending excessive time reading irrelevant and unproductive content, etc. These problems can be partially overcome by employing personalization (the Adaptive Web), semantics (the Semantic Web) or social relationships (the Social Web). We focus on the user modeling area and we exploit existing knowledge, mostly from the Adaptive and Semantic Web. User model in the adaptive web-based applications consists of user’s characteristics that are used for personalization of layout, navigation or content. There are methods that use explicit or implicit approaches, or their combination to acquire user characteristics and keep them up to date. The result is that accurate personalization can be provided to a user.

We present a contribution to the current state of the art in the user modeling area, namely we focus on creation and maintenance of the user model.

Results

We proposed three novel methods for acquisition and maintenance of user characteristics in the user model.

  • The first method is based on generating questions to be used for user model. Particular questions are generated according to the attributes of information concepts that are the subject of the specific application domain. The entire process of asking questions is driven by user defined rules.
            The method was evaluated using the job offer (research project NAZOU) and scientific publications (research project MAPEKUS) domain ontology using our software prototype ExACT (Explicit Actualizer). The templates were proposed with regard to their universality. We designed 7 templates for the job offer domain. These templates allow generating questions for 22 characteristics. For the scientific publications domain we proposed 3 templates which allow generating 5 questions. The small amount of templates (questions) in this domain is influenced by straightforward structure of the domain.

  • The second method is based on the content analysis and assumes that comparing attributes of documents, which were found interesting for a user, can be a source for discovering information about user’s interests. We use ontology structure and different similarity metrics to compute similarity between instances of ontological concepts. Moreover, we impute reasons that might have caused user’s interest in the content.
            We have developed the software tool ConCom (Concept Comparer) that automates the proposed method. ConCom supports the computing of two kinds of similarity — either for all properties or only for properties that are common for both compared instances. Our experiments showed that similarity where all properties are considered is more suitable for discovering user characteristics, while similarity computed for common properties only better mimics the similarity estimated by real users. The tool was incorporated into the pilot application that was developed in the course of the research project NAZOU.

  • The third and last method is based on spreading activation. If there are connections between information concepts (e.g., learning objects in an educational application) of the domain model user’s characteristics can be utilized even for concepts that have not been visited yet. In such a way, more accurate and responsive information retrieval capabilities for the user become available.
            We described our extension of domain model that enables more accurate adaptation using two related parts — knowledge item spaces and learning object spaces — that can be reused across several educational applications. We developed a prototype of an adaptive web-based educational application that recommends learning objects from the domain model. The prototype was implemented in the course of the PeWePro research project. The course consisted of 16 learning objects containing simple text, exercise and explanation types. Knowledge items, which learning objects are assigned to, came from ACM classification. Knowledge item space contained 1 476 topics and 4 keywords added especially for our course. Benefit of this approach is in changing user’s characteristics in the entire domain model and not only in the parts that the user has already visited.

Conclusion

Main contribution of this work is a proposal of three novel methods to automatic acquisition and maintenance of user characteristics that employ semantics provided by ontological representation. Another contribution of this work is its aim at the methods for acquisition and maintenance of user characteristics. Currently, there is less attention paid to these problems (in comparison to adaptive navigation and presentation) in the adaptive hypermedia field. One of the reasons is that changes of the user model are rather considered as a support for the personalization and are not directly visible on presentation layer.

Selected publications

A. Andrejko
Novel Approaches to Acquisition and Maintenance of User Model. Dissertation thesis, Slovak University of Technology in Bratislava 2009. pdf
A. Andrejko, M. Bieliková
Comparing instances of ontological concepts for personalized recommendation in large information spaces. Computing and Informatics. To appear.
A. Andrejko, M. Bieliková
Personalized Comparing Instances of Domain Ontology Concepts. In Phivos Mylonas, Manolis Wallace, and Marios Angelides (Eds.): Proceedings of SMAP 2008 — 3rd International Workshop on Semantic Media Adaptation and Personalization, Prague, Czech Republic, CS IEEE Press, 2008, pp. 76–81.
A. Andrejko, M. Bieliková
Investigating Similarity of Ontology Instances and its Causes. In Vera Kurkova, Roman Neruda, and Jan Koutnik (Eds.): Artificial Neural Networks — ICANN 2008, LNCS 5164, Prague, Czech Republic, Springer, 2008, pp. 1–10.
M. Šimún, A. Andrejko, M. Bieliková
Maintenance of Learner’s Characteristics by Spreading a Change. In Learning to Live in the Knowledge Society, volume 281 of IFIP International Federation for Information Processing, Springer Boston, 2008, pp. 223–226.
A. Andrejko, M. Barla, M. Bieliková
Ontology-based User Modeling for Web-Based Information Systems. In Wita Wojtkowski, W. Gregory Wojtkowski, Jože Zupancic, Gabor Magyar, and Gabor Knapp (Eds.): Advances in Information Systems Development, New Methods and Practice for the Networked Society, volume 2, Springer Science+Business Media, New York, 2007, pp. 457–468.
A. Andrejko, M. Barla, M. Bieliková, M. Tvarožek
User characteristics acquisition from logs with semantics. In ISIM 2007 Information Systems and Formal Models: 10th International Conference on Information System Implementation and Modeling, 2007, pp. 103–110.
M. Šimún, A. Andrejko, M. Bieliková
Ontology-based Models for Personalized E-Leraning Environment. In ICETA 2007: 5th Internation Conference on Emerging e-Learning Technologies and Applications, Elfa, Stará Lesná, Slovak Republic, 2007, pp. 335–340.

to Homepage to Teaching to the Top

Home
Research
Projects
Publications
Books
SCM
Teaching
Links
Last updated:
Mária Bieliková bielik [zavináč] fiit-dot-stuba-dot-sk
Design © 2oo1 KoXo