Program | Abstracts

Birger Larsen (Information Interaction and Information Architecture Department, Royal School of Library and Information Science, Copenhagen, Denmark)
    Polyrepresentation 2009 - Principle or Theory?

    The idea of polyrepresentation was proposed by Peter Ingwersen in 1994/1996 as one of several consequences of a cognitive view on interactive information retrieval. In brief, polyrepresentation puts emphasis on the diversity resulting from the different cognitive actors in involved in IIR, and proposes to exploit this to achieve better performance, e.g., data fusion. Originally called a 'theory' polyrepresentation has later been referred to as a 'principle' as it did not fulfill an important criterion to be reckoned as a theory (at least according to some philosophies of science, cf. Bunge, 1967): that of having its predictions supported by sufficiently strong empirical evidence. The talk will outline the idea of polyrepresentation, review the empirical evidence published to date and examine its current status as principle or theory.

Robert Jäschke (Department of Knowledge and Data Engineering, Universität Kassel, Kassel, Germany)
    Tag Recommendations in BibSonomy

    Since the emergence of collaborative tagging systems, research in the field of tag recommendations has established evaluation protocols, defined baselines, and came up with quite some valuable methods. In this talk we give an overview of our work in the field of tag recommendations and present insights we gathered by organizing this year's ECML PKDD Discovery Challenge. Furthermore, our collaborative tagging system BibSonomy will be presented as platform for organizing scientific articles as well as for testing and evaluating knowledge discovery methods. We conclude with an outlook on the future of BibSonomy and on upcoming research challenges.

Frank Hofstede (Search Expertise Centrum, Utrecht, The Netherlands)
    Recommendation Orchestration

    Recommender systems are made for their users and serve their users needs. Developing recommender systems leaves many options open to the researcher. Is it wise to leave the decision on what to research (and therefore decide on the direction of your research) to the researcher?

Maarten de Rijke (Informatics Institute, University of Amsterdam, Amsterdam, The Netherlands)
    Open Sources for Expertise Retrieval

    Modern expertise retrieval systems can use mixtures of structured data and full-text documents linked to a person to compute associations between people and topics. These associations can then be used to rank people with respect to a topic and topics with respect to a given individual. In this talk I will show that the additional use of open sources ("the web") for expertise retrieval leads to highly significant improvements in expertise retrieval effectiveness.

    In addition to an example-driven analysis of the effect of using open sources, I will touch on two open issues: (i) the choice of optimal model for integrating evidence from open sources turns out to be highly topic dependent, and (ii) when using a mixture of open sources and other types of evidence, it is not clear how to identify suitable extractive "result snippets" that provide convincing evidence in support of an association between a person and a topic.

    This is based on joint work with Krisztian Balog.


                           

Last update: