=Paper=
{{Paper
|id=None
|storemode=property
|title=The Return of the Probability of Relevance
|pdfUrl=https://ceur-ws.org/Vol-986/paper_0.pdf
|volume=Vol-986
|dblpUrl=https://dblp.org/rec/conf/dir/Fuhr13
}}
==The Return of the Probability of Relevance==
https://ceur-ws.org/Vol-986/paper_0.pdf
The Return of the Probability of Relevance
Norbert Fuhr
The University of Duisburg-Essen
Duisburg, Germany
norbert.fuhr@uni-due.de
Abstract About the Speaker
The probability ranking principle (PRP) proves that rank- Dr. Norbert Fuhr is a full professor in the Department of
ing documents by decreasing probability of relevance yields Computer Science at the University of Duisburg-Essen. He
optimum retrieval quality. Most research on probabilistic obtained his Ph.D in Computer Science from the Technical
models has focused only on producing a probabilistic rank- University of Darmstadt in 1986 where he served as an assis-
ing, without estimating the actual probabilities. In this talk, tant professor. He became Associate Professor in the com-
we discuss models for three types of modern IR applications puter science department of the University of Dortmund in
which rely on calibrated values of the probability of rele- 1991, before taking up his current position in 2002.
vance. He has published more than 300 papers in the fields of IR,
databases and digital libraries. His current research interests
are retrieval models, networked digital library architectures,
1. Vertical search deals with the aggregation of docu-
user-oriented retrieval methods and the evaluation of digital
ments with different types or media (such as, e.g.,
libraries.
Web pages, news, tweets, videos, images) in response
He has served as regular PC member of many major interna-
to a query. Based on the probabilistic estimation of
tional conferences related to information retrieval and digital
the number of relevant documents per resource, the
libraries, such as ACM-SIGIR, CIKM, ECIR, SPIRE, ICDL,
decision-theoretic selection model describes the opti-
ECDL, ICADL, FQAS. He was PC chair of ECIR 2002, IR
mum solution for this problem.
track chair of CIKM 2005 and Co-Chair of SIGIR 2007. For
2. The optimum clustering framework provides not only the German IR-group GI-FGIR, he served as Chair from
the first theoretic foundation for document clustering, 1992-2008. He also is a member of the editorial boards of the
it also proves the clustering hypothesis. Its key idea journals Information Retrieval, ACM Transactions on Infor-
is to base cluster analysis and evaluation on a set of mation Systems, International Journal of Digital Libraries,
queries, by defining documents as being similar if they and Foundations and Trends in Information Retrieval.
are relevant to the same queries. In 2012, he received the prestigious Gerald Salton Award in
recognition of his significant, sustained and continuing con-
3. The interactive PRP generalizes the classical PRP for tributions to research in information retrieval.
interactive retrieval. It characterizes each situation in The committee particularly emphasised his ”pioneering con-
interactive retrieval as a list of choices, where each tributions to the theoretical foundations of information re-
choice is described as the effort for evaluating it, the trieval and database systems. His work describing how learn-
probability that the user will accept it, and the benefit ing methods can be used with retrieval models and index-
resulting from acceptance. By developing appropriate ing anticipated the current interest in learning ranking func-
parameter estimation methods, we can describe inter- tions, his development of probabilistic retrieval models for
active retrieval by Markov models, which allow for a database systems and XML was ground-breaking, and his
number of predictions. recent work on retrieval models for interactive retrieval has
inspired new research. His rigorous approach to research and
With these models, it becomes possible to implement ap- research methods is an outstanding example for our field.”
proaches based on solid theoretic foundations, which are
more transparent than heuristic approaches, thus allowing
for theory-guided adaptation and tuning.