Getting docs that have match on a search phrase is easy (using case-,
punctuation-, white-space-, insensitive options), and finding docs that have
the highest frequency for the words in the search phrase is easy
(cts:word-query and a sequence of terms), but I want to find docs that most
closely match the search phrase.
For example, if I have a doc that has this text in it: "Mary had a little lamb
whose fleece was white as snow"
If I search using "mary had a little lamb whose fleece was white as SNOW!!!" a
cts:word-query would match if I sent the entire phrase and used all the
"insensitive" options.
If I search by tokenizing the phrase into ("mary", "had", "little", "lamb",
"fleece", "white", "snow") I will get the doc that has the highest frequency of
those words (and weighted according to doc size), which may or may not be my
"Mary had a little lamb doc".
And if I search for "Jane had a little lamb whose fleece was white as snow" the
Mary doc won't match because the phrase doesn't match, and a tokenized words
search probably won't match because some other doc with "Jane" and "snow" or
whatever would be higher priority. I can try to use a near query of all the
words except "Jane" isn't in the doc so there's be no match for my Mary doc.
What I want is the doc that has a phrase that most closely matches the search
phrase, even if I drop, replace, or introduce an incorrect word. And I mean
more than just spelled wrong.
You can see that "Jane had a little lamb whose fleece was white as snow" is
really close to "Mary had a little lamb whose fleece was white as snow" but I
can't quite figure out how to get MarkLogic to determine that quickly since the
phrase won't match and tokenized words won't necessarily give me the best
relevance. I can get all the permutations of the phrase (every word with all
the other words in all combinations) and OR them together but search
performance suffers after just a few permutations.
Anyone know how to do this?
thanks,
-Ryan
_______________________________________________
General mailing list
General@developer.marklogic.com
http://community.marklogic.com/mailman/listinfo/general
