=Paper=
{{Paper
|id=Vol-1167/CLEF2001wn-adhoc-vanderWeerdEt2001
|storemode=property
|title=First Experiments with CLEF
|pdfUrl=https://ceur-ws.org/Vol-1167/CLEF2001wn-adhoc-vanderWeerdEt2001.pdf
|volume=Vol-1167
|dblpUrl=https://dblp.org/rec/conf/clef/WeerdB01
}}
==First Experiments with CLEF==
<pdf width="1500px">https://ceur-ws.org/Vol-1167/CLEF2001wn-adhoc-vanderWeerdEt2001.pdf</pdf>
<pre>
                    First experiments with CLEF
                                    Peter van der Weerd
                                    pweerd@medialab.nl

                                       Wilfred Blom
                                    wblom@medialab.nl

                       Medialab BV, Schellinkhout, The Netherlands
                                 http://www.medialab.nl



1. Goal
Our goal for participation in CLEF was to compare our information-retrieval engine with
other engines, and to get more information about other approaches to the information
retrieval problems. The Olympic spirit dominated MediaLab’s first participation. To play is
more important than to win.


2. Preparation
Due to the very limited time frame we had, we decided to use a straightforward approach. We
decided to use our standard retrieval engine without any modifications, and to only parse the
supplied queries by breaking them into words and feed the individual words into the retrieval
engine.

Before we started the experiment it was recognized that our score would not be very large:
the queries specified rather specific what to return and what not. Because we used all words
from the query to find our results we knew on forehand that we would find the undesired
results as well. However, we felt that it was nice to check out if we could deliver better
results than the worst participants with this approach.

Two indexes were built:
      • A title field consisting of the concatenation of the shorter fields: the TI, LE and
          OS fields from the articles.
      • A body field consisting of the TE-field


3. Query processing
Besides normal search behavior we used the following add-ons to process the queries:
       • Stop words
           We used the stop words supplied by the CLEF project
       • Stemming.
           The built-in Dutch stemmer is based on the porter stemmer
       • Compound term searching.
           MediaLab’s search engine has built-in facilities to expand for example “roos” to
           “klaproos”.
All fields from the query (<NL-title>, <NL-desc>, <NL-narr>) were stripped into words,
sending each word into the engine, giving a ranked set of documents for each word. These
individual results were combined using a bounded-add of their weights:
                                 w = w1 + w2 – w1*w2

This final result was submitted to CLEF.


4. Results and future work
As one might expect the resulting precision was not very high: 22% of the queries scored
above the median, the rest below. However, our results never were the worst! Realizing that
the results of machine-driven approaches as we did and (a few) manual-driven queries are
intermixed within the results, we are not at all discontented with the results.

The first thing we will do is to incorporate NLP into the query-process, and to use weighted
queries. MediaLab is enthusiastic about CLEF and intends to allocate more resources for next
year’s participation.


5. Acknowledgements
The participation of MediaLab to CLEF would not be realized without the unlimited
enthusiasm and push of Dennis Reidsma who visited us this year to get his degree at the
University of Twente.

</pre>