|
The metasearcher on this web site is offered as a search
instrument and research tool. As search instrument, it offers
improved coverage of the nearly one billion pages on the Web by
allowing the user to compile results from a collection of search
engines. As research tool, it provides the means to assess the
performance of search engines.
Performance is assessed by comparing one search engine with a
collection of search engines. A set of search engines (possibly
including the one under investigation) is used to establish a
benchmark. Then the performance of a particular search engine is
compared with the benchmark. The comparison is a measure of the
dissimilarity between a vector representing the benchmark and one
representing the performance of a particular search engine. Three
different metrics are used to capture the distance. The first
ignores the order of presentation of URLs retrieved in response to
a query. The second and third take order into account. A report
detailing these measures is in preparation. If interested, write
us at akira@cs.ccny.cuny.edu.
Instructions for Use
- Select a subset of search engines from the menu on the left. The
process of selecting multiple engines varies with the type of machine
used to connect to the Internet. For example, using PCs, one holds
down the control key while clicking the engine name. If you want to
use all of the available search engines, click **Select
All**. Search engines whose names are enclosed in parentheses
are not currently available.
- Specify search terms, one per line, in the text box on the right.
Any number of terms can be used in one search. A search term can be a
word, phrase or boolean combination of words and phrases. If one
wants to search on the exact text of a term, enclose the text in
double quotation marks. For more details, inspect tips for advanced
search provided by the individual search engines selected for your
search.
- The number of responses per search engine can be specified up to
maximum of 50.
- Two URLs are treated as identical if they are identical as
character strings. Some URLs correspond to Web sites with many
different pages. If you wish to distinguish all the pages associated
with the Web sites for the purpose of performance measurement, then
choose the "Complete URL" option. Otherwise (to distinguish Web sites
only) select the "Truncated URL" option.
- 'Prominence Summary' presents a list of URLs in the order of
their frequency of occurrence among the response sets. 'Weightings'
determine the increment that is added for each occurrence of a URL.
Interpretation of Results
- Two types of output are presented:
- Bias metrics
- Response sets of URLs
- Bias metrics are computed cumulatively for the search terms. The
first set of metrics is based on the response sets retrieved for the
first search term. The second set is based on the union of the response
sets corresponding to the first and second search terms, and so on.
- Each search term generates a collection of response sets, one
for each search engine selected. The collections are listed in order
of search terms.
|
