Getting fuzzy match information

["Allan, Brad (Bracknell)" <Br...@Fiserv.com>]

Has anyone done or know of work done that would help me to get detailed information about my hits with regard to fuzzy matches? Also very happy to receive suggestions :).

I'm looking to obtain the similarity percentage of each token in the each hit.

Example: fuzzy query looks something like this:
(name:80% similar to "john" or name:80% similar to "henry" or name:80% similar to "smith")
And I get hits:

*         Jon George Smythe

*         John Joe Henry

*         Smith John & Carter engineering
All valid hits, however my users want to be able to view the similarity and indeed prioritise certain actions by being able to compare the results of 2 different searches (and therefore normalised scores are not as useful as knowing the actual similarity information).

Clearly this sort of ability does not make sense when one is searching in large amounts of data (documents), but in my case I'm searching through a set of names and some additional person information.

Options could be to post process the hits and use/lift the FuzzyTermEnum logic to re-compute the similarity value. Or perhaps extend the FuzzyQuery to register a 'listener' to receive the information?
Other ideas? Thoughts?



________________________________

CheckFree Solutions Limited (trading as Fiserv)
Registered Office: Eversheds House, 70 Great Bridgewater Street, Manchester, M15 ES
Registered in England: No. 2694333

Re: Getting fuzzy match information

[Simon Svensson <si...@devhost.se>]

Hi,

I feel that http://wiki.apache.org/lucene-java/ScoresAsPercentages is 
somewhat relevant, even if you're not strictly talking about normalizing 
your scores.

I would look into calling indexSearcher.Explain(Query query, Int32 
docId) to retrieve an Explanation object that _should_ container 
information how your document matched, including scores for every term. 
(The actual user interface for presenting this is left to the reader. 
Moahahahaa!)

// Simon

On 12/12/13 17:22, Allan, Brad (Bracknell) wrote:
> Has anyone done or know of work done that would help me to get detailed information about my hits with regard to fuzzy matches? Also very happy to receive suggestions :).
>
> I'm looking to obtain the similarity percentage of each token in the each hit.
>
> Example: fuzzy query looks something like this:
> (name:80% similar to "john" or name:80% similar to "henry" or name:80% similar to "smith")
> And I get hits:
>
> *         Jon George Smythe
>
> *         John Joe Henry
>
> *         Smith John & Carter engineering
> All valid hits, however my users want to be able to view the similarity and indeed prioritise certain actions by being able to compare the results of 2 different searches (and therefore normalised scores are not as useful as knowing the actual similarity information).
>
> Clearly this sort of ability does not make sense when one is searching in large amounts of data (documents), but in my case I'm searching through a set of names and some additional person information.
>
> Options could be to post process the hits and use/lift the FuzzyTermEnum logic to re-compute the similarity value. Or perhaps extend the FuzzyQuery to register a 'listener' to receive the information?
> Other ideas? Thoughts?
>
>
>
> ________________________________
>
> CheckFree Solutions Limited (trading as Fiserv)
> Registered Office: Eversheds House, 70 Great Bridgewater Street, Manchester, M15 ES
> Registered in England: No. 2694333
>