Thursday, June 21, 2007

Background (4): Web Personalization and User Profile


Personalization mechanisms in literature can be divided to three categories. These mechanisms try to predict user interest in a particular item [2].
Demographic: similarity of current item properties with items that users liked in the past [1].
Content-based: based on the similar properties of the items that user liked in the past [1].
Collaborative: based on the rating patterns of similar users (the choices of people that liked similar objects as the current users are recommended) [3].

Advantages and disadvantages of these mechanisms are discussed in [4]. Demographic filtering (recommended) is more adaptable to the preference changes comparing to content-based filtering but requires some information which sometimes user is not willing to provide [2]. Collaborative filtering is a good alternative to demographic filtering, as it does not rely on information about the users and the items.


Reference
[1] C. Basu, H. Hirsh, W. Cohen, Recommendation as classification: using social and content-based information in recommendation, in: Proceedings of AAAI-98, Menlo Park, CA, USA, 1998, pp. 714–720.

[2] Stegers R., Fekkes P. and StuckenschmidtH. MusiDB: A personalized search engine for music, Journal of Web Semantics: Science, Services and Agents on the World Wide Web, 2006, Pp 267-275

[3] D. Goldberg, D. Nichols, M. Oki Douglas Terry, Using collaborative filtering to weave an information tapestry, in: Communications of the ACM, vol. 35, issue 12, ACM Press, New York, USA, 1992, pp. 61–70


[4] R. Burke, Hybrid recommender systems: survey and experiments, in: User Modeling and User-Adapted Interaction, vol. 12, issue 4, Springer, 2002, pp. 331–370.

Wednesday, June 6, 2007

Background (3) Web Personalization

The web is a huge information repository and finding relevant information in this environment is not a trivial task. Web personalization aims to help users find relevant information and services efficiently. The main issue is that the profile of the user must be recognized by the web server to provide him personalized services. Different approaches are proposed to overcome this problem. Current approaches could be divided to server-side accounts, cookies, and identity profiles (e.g Microsoft password). The disadvantage of server-side accounts is that the user should enter the same information in different websites. In addition you should remember lots of username and passwords. The problem with cookies is that they are based on the server technology which has a different standard and coding from one web server to the other, thus they are not applicable for different services. Also cookies are not meaningful for the user. Although identity profile can handle a few services that are using the same standard at a time but again it is not applicable to other services on the web. To summarize the above, current approaches are incapable of using and integrated information of the user for different services. Privacy and security is another issue in the current mechanisms. Semantic web introduces an architecture that is suitable for web personalization. There are different mechanisms using semantic web concepts for web personalization and user profiling. Ontologies were proved as a handy mean to represent user profiles and preferences. [1] introduces an extension of the GET method in HTTP to include a new parameter that points to the URL of the user’s FOAF [2] file. FOAF files are easy to understand and based on an open standard format. In this way the web server can understand the user preferences using the FOAF file. The user profile unlike the user-centric identity management is portable and can be accessed on the web by different web servers. Baoyao et. al. introduces a new web usage mining [4] approach to model web access behavior of users based on discovered user access patterns from client-side access logs [3]. This model is transformed to an ontology and can be used to provide personalized web services to the user. The ontology is generated using Formal Conceptual Analysis [5] based on fuzzy logic. [6] exploits ontologies with fuzzy relations to represent user profiles. This ontology-based personalization is very helpful for complex retrieval tasks in multimedia domain. It enhances RDF with novel characteristics and the proposed model is a graph with concepts as nodes, and the edge between two nodes that forms a contextual relation between concepts. Reference

Reference
[1] Ankolekar A., Varandecic D. Personalizing web surfing with semantically enriched personal profiles. In Makram Bouzid and Nicola Henze, Proceedings of the Semantic Web Personalization Workshop. Budva, Montenegro, June 2006.

[2] http://www.foaf-project.org/

[3] Zhou B., Hui C. S., and Fong A. Web Mining Research: A Survey. In: ACM SIGKDD Explorations, 2 (2000) 1-15.

[5] Stumme G., and Maedche A., Ontology Merging for Federated Ontologies on the Semantic Web. In: Workshop on Ontologies and Information Sharing, at IJCAI, Seattle, USA , (2001).

[6] Ph. Mylonas, D. Vallet, M. Fernández, P. Castells and Y. Avrithis. Ontology-based Personalization for Multimedia Content. 3rd European Semantic Web Conference - Semantic Web Personalization Workshop, Budva, Montenegro, 11-14 June 2006

Monday, June 4, 2007

Summary of Background 1 and 2

Below is a picture from the white board I drew to summarizes the previous Background part 1 and 2 in a presentation. This is trying to bring all recent attempts regarding utilizing approximate reasoning in the Semantic Web in a glance.