My formula was clear, but oops it was wrong. Let's try again. Here's a quote from engineering about confidence (in which I have high confidence!):
-- I believe a match would have to be the only match out of a large database to get a confidence of one. Here's how I think of confidence. Imagine yourself a librarian, and someone walks up to you and says "I would like a book about X". You can almost always give them some book that would match what they asked for, but sometimes you are going to be more confident about whether the book is the one they really were looking for than other times. For example, if someone asked for a book about famous left-handed vice presidents, and happen to have a book called "Famous Left-Handed Vice-Presidents", you would be pretty confident that they book is the one they were looking for. If someone asked for a blue book with words in it, you could still give them a blue book with words in it, but you would not be very confident that it was the book they were looking for. -- Both confidence and fitness involve lengthy math formulas. If you really want to know how the numbers are calculated, contact me privately and I'll give you some tips. -jh- On May 4, 2011, at 8:31 AM, Ton Long Hiep wrote: > Dear Jason Hunter, > > Thank you very much for your information. > First, about confidence score, your formula is clear. However, I have just > done a search on Marklogic DB and it returns only one result: > <search:response total="1" start="1" page-length="10" xmlns="" > xmlns:search="http://marklogic.com/appservices/search";> > <search:result index="1" uri="...xml" path="fn:doc("...xml")" > score="92" confidence="0.3111" fitness="0.408157"> > <search:snippet> > ... > </search:snippet> > </search:result> > > In this case, applying your formula, confidence score should be sqrt(92/92) = > 1. However, actually it is 0.3111? Could you explain it for me? > Second, regarding to fitness score, why is it "0.408157" in this case? Could > you tell me? > > I am looking forwards to hearing from you. > > Thank you very much! > Long Hiep. > > On Sun, May 1, 2011 at 11:45 PM, Jason Hunter <[email protected]> wrote: > On Apr 26, 2011, at 4:54 AM, Ton Long Hiep wrote: >> As I read in the document "Search application developer's guide", the >> confidence score is defined as follows: >> "Confidence scores are calculated by first bounding the scores between 0 and >> 1.0, and then taking the square root of the bounded number" >> However, there is no formula how it is calculated. Could someone help me? > > Lets say you have results with scores up to 500. Well, the confidence of the > top result item will be 1.0. The confidence of a result item with score 100 > will be sqrt(100/500) = 0.4. > > There's no top score, so confidence is a way to normalize and bound the > unbounded. > > Think of confidence as a way to quantify how much better one result item is > than another result item. > >> Regarding to fitness score, it is stated as follows: >> "The fitness number does not include any quality settings that might be on >> the document, and it does not use document frequency in the calculation" >> However, when I do search, the keyword which I search exists in the document >> but the document's fitness is still below 1. Could someone explain it for me? > > Fitness also normalizes between 0.0 and 1.0 but ignores quality (the inherent > goodness value you can assign to a document) and it ignores IDF (inverse > document frequency, which is used so more rare terms have higher impact on > score). It's just a report on how well the given document matches the given > query. > > The values will always be between 0.0 and 1.0 because the results are > designed to "normalize" all values to that range. > > -jh- > > > _______________________________________________ > General mailing list > [email protected] > http://developer.marklogic.com/mailman/listinfo/general > > > _______________________________________________ > General mailing list > [email protected] > http://developer.marklogic.com/mailman/listinfo/general
_______________________________________________ General mailing list [email protected] http://developer.marklogic.com/mailman/listinfo/general
