Thursday, September 4, 2008

Follow up to precision/recall curve question

This is just a follow up to the mail I sent regarding the plotting of precision/recall curves…I realize that the example in the slides shows 10 relevant documents and it may seem as if this is the reason for plotting the 11 point curve however, this is not the case…

 

11 point curves are frequently used irrespective of the number of relevant documents…for some additional information please refer to these links, send me a mail, or stop by and see me

 

http://datamin.ubbcluj.ro/wiki/index.php/Evaluation_methods_in_text_categorization#11-point_average_precision

 

http://nlp.stanford.edu/IR-book/html/htmledition/evaluation-of-ranked-retrieval-results-1.html

 

Garrett

 

 

1 comment:

Garrett Wolf said...

Just a follow up...

It was asked why the 11-point precision/recall curve is used instead of simply plotting a point after each relevant document is retrieved....

One reason for this is due to the fact that often times we are not plotting the precision/recall curve for a single query, but rather a set of queries averaged. Given that each query may have differing number of relevant documents, it's useful to measure the precision for each query at consistent intervals.