You are viewing a plain text version of this content. The canonical link for it is here.
Posted to java-user@lucene.apache.org by Amr ElAdawy <Am...@etisalat.com> on 2011/01/03 08:15:39 UTC
Search Score percentage, Should not be relative to the highest
score
Dear All,
When using lucene to search documents, the results have a
score based on their relativity to the search term. Inside lucene, the score
percentage is calculated as a percentage of the maximum score achieved.
Assume we are searching for Amr ElAdawy
We get results like the following along with the score:
Omar ElAdawy --> 1.6
Amro ElAdawy --> 1.9
Amir Adawi --> 1.3
With no exact match. Then the percentage will be calculated relative to the
highest score ( despite the fact that it is not exact match). So the
percentage will be:
Omar ElAdawy --> 84.21%
Amro ElAdawy --> 100%
Amir Adawi --> 68.42%
I need to change that, so the percentage will be relative to the exact match
score. And will be something like that (assuming the Matching result score
is 3):
Omar ElAdawy --> 53.33%
Amro ElAdawy --> 63.33%
Amir Adawi --> 43.33%
using lucene 2.9.4
Regards,
Amr ElAdawy
--
View this message in context: http://lucene.472066.n3.nabble.com/Search-Score-percentage-Should-not-be-relative-to-the-highest-score-tp2183420p2183420.html
Sent from the Lucene - Java Users mailing list archive at Nabble.com.
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org
Re: Search Score percentage, Should not be relative to the highest
score
Posted by Amr ElAdawy <Am...@etisalat.com>.
any ideas?
--
View this message in context: http://lucene.472066.n3.nabble.com/Search-Score-percentage-Should-not-be-relative-to-the-highest-score-tp2183420p2265633.html
Sent from the Lucene - Java Users mailing list archive at Nabble.com.
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org
Re: Search Score percentage, Should not be relative to the highest
score
Posted by Amr ElAdawy <Am...@etisalat.com>.
2.9.4
--
View this message in context: http://lucene.472066.n3.nabble.com/Search-Score-percentage-Should-not-be-relative-to-the-highest-score-tp2183420p2199732.html
Sent from the Lucene - Java Users mailing list archive at Nabble.com.
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org
Re: Search Score percentage, Should not be relative to the highest score
Posted by Ahmet Arslan <io...@yahoo.com>.
> Did not work,
>
> I am using my own Similarity and the coord method is not
> called, because the
> disableCoord variable is set to true from FuzzyQuery
>
>
> public Similarity getSimilarity(Searcher searcher) {
> Similarity result =
> super.getSimilarity(searcher);
> if (disableCoord) {
>
> // disable coord as
> requested
> result = new
> SimilarityDelegator(result) {
> public float coord(int
> overlap, int maxOverlap) {
> return 1.0f;
> }
> };
> }
> return result;
> }
What is the version of Lucene you are using?
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org
Re: Search Score percentage, Should not be relative to the highest
score
Posted by Amr ElAdawy <Am...@etisalat.com>.
Did not work,
I am using my own Similarity and the coord method is not called, because the
disableCoord variable is set to true from FuzzyQuery
public Similarity getSimilarity(Searcher searcher) {
Similarity result = super.getSimilarity(searcher);
if (disableCoord) { // disable coord as
requested
result = new SimilarityDelegator(result) {
public float coord(int overlap, int maxOverlap) {
return 1.0f;
}
};
}
return result;
}
any ideas.
--
View this message in context: http://lucene.472066.n3.nabble.com/Search-Score-percentage-Should-not-be-relative-to-the-highest-score-tp2183420p2197077.html
Sent from the Lucene - Java Users mailing list archive at Nabble.com.
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org
Re: Search Score percentage, Should not be relative to the highest
score
Posted by Amr ElAdawy <Am...@etisalat.com>.
I am overriding the coord method in my customSimilairty Class and it will be
return (float)overlap/(float)maxOverlap;
I'll update you.
Thanks for your help
--
View this message in context: http://lucene.472066.n3.nabble.com/Search-Score-percentage-Should-not-be-relative-to-the-highest-score-tp2183420p2192261.html
Sent from the Lucene - Java Users mailing list archive at Nabble.com.
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org
Re: Search Score percentage, Should not be relative to the highest score
Posted by Ahmet Arslan <io...@yahoo.com>.
So, can we say that if you have something that gives you the "how many query terms matched" info, will that satisfy your requirement?
Query: term1 term2
Doc1: term1 term2 => n=2 => %100
Doc2: term1 term2 term3 term4 => n=2 => %100
Doc3: term1 term1 term3 => n=1 => %50
Doc4: term2 term3 term4 => n=1 => %50
If yes Explanation will you give that info in coord part. For example coord(1/3) means one query term matched and there are total 3 query terms.
Here is an example Explanation:
0.013397463 = (MATCH) product of:
0.040192388 = (MATCH) sum of:
0.040192388 = (MATCH) weight(pagetext:para in 34930), product of:
0.46250778 = queryWeight(pagetext:para), product of:
3.1780937 = idf(docFreq=5546, maxDocs=48977)
0.14552994 = queryNorm
0.086901 = (MATCH) fieldWeight(pagetext:para in 34930), product of:
1.0 = tf(termFreq(pagetext:para)=1)
3.1780937 = idf(docFreq=5546, maxDocs=48977)
0.02734375 = fieldNorm(field=pagetext, doc=34930)
0.33333334 = coord(1/3)
--- On Mon, 1/3/11, Amr ElAdawy <Am...@etisalat.com> wrote:
> From: Amr ElAdawy <Am...@etisalat.com>
> Subject: Re: Search Score percentage, Should not be relative to the highest score
> To: java-user@lucene.apache.org
> Date: Monday, January 3, 2011, 3:09 PM
>
> Consider the following.
>
> Query: term1 term2
> Doc1: term1 term2
> Doc2: term1 term2 term3 term4
> Doc3: term1 term1 term3
> Doc4: term3 term4
>
> For the above documents, Doc1 and Doc2 will b exact match (
> as they contain
> all the terms in the search Query). Doc3 is partially match
> as it contains
> term1 only (we neglect the term frequency tf always 1
>
>
> The score percentage ( calculated by Lucene in Hits.java
> line 133) and will
> be
>
> Doc1: 100%
> Doc2: 100%
> Doc3: 80%
>
> This is not a problem at all, the problem occurs when there
> is no exact
> matching document as following:
>
> Query: term1 term2
> Doc1: term1 term3
> Doc2: term2 term3 term4
> Doc3: term1 term1 term3
> Doc4: term3 term4
>
>
> The score will be calculated as
>
> Doc1: 100%
> Doc2: 100%
> Doc3: 50%
>
> You can see that Doc1 and Doc2 got 100% despite that they
> are not exact
> match. but as they got the highest score, Lucene considers
> them 100% match.
>
> This is my problem
>
> All I need is to make the percentage correct in the second
> case so it will
> be something as
>
> Doc1: 50%
> Doc2: 50%
> Doc3: 30%
>
> I hope I made myself clear.
>
>
> --
> View this message in context: http://lucene.472066.n3.nabble.com/Search-Score-percentage-Should-not-be-relative-to-the-highest-score-tp2183420p2184613.html
> Sent from the Lucene - Java Users mailing list archive at
> Nabble.com.
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-user-help@lucene.apache.org
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org
Re: Search Score percentage, Should not be relative to the highest
score
Posted by Amr ElAdawy <Am...@etisalat.com>.
Consider the following.
Query: term1 term2
Doc1: term1 term2
Doc2: term1 term2 term3 term4
Doc3: term1 term1 term3
Doc4: term3 term4
For the above documents, Doc1 and Doc2 will b exact match ( as they contain
all the terms in the search Query). Doc3 is partially match as it contains
term1 only (we neglect the term frequency tf always 1
The score percentage ( calculated by Lucene in Hits.java line 133) and will
be
Doc1: 100%
Doc2: 100%
Doc3: 80%
This is not a problem at all, the problem occurs when there is no exact
matching document as following:
Query: term1 term2
Doc1: term1 term3
Doc2: term2 term3 term4
Doc3: term1 term1 term3
Doc4: term3 term4
The score will be calculated as
Doc1: 100%
Doc2: 100%
Doc3: 50%
You can see that Doc1 and Doc2 got 100% despite that they are not exact
match. but as they got the highest score, Lucene considers them 100% match.
This is my problem
All I need is to make the percentage correct in the second case so it will
be something as
Doc1: 50%
Doc2: 50%
Doc3: 30%
I hope I made myself clear.
--
View this message in context: http://lucene.472066.n3.nabble.com/Search-Score-percentage-Should-not-be-relative-to-the-highest-score-tp2183420p2184613.html
Sent from the Lucene - Java Users mailing list archive at Nabble.com.
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org
Re: Search Score percentage, Should not be relative to the highest score
Posted by Ahmet Arslan <io...@yahoo.com>.
> I had read the link and I understand the concern, however,
> the normalization
> is happening inside lucene. Where the normalizing value is
> the inverse of
> the maxScore.
>
> I can alter the code to leave the original score, however
> it is a business
> requirements to view the matching percentage. Also, the
> absolute score means
> nothing because we wont know the exact matching score.
Can you re-phrase your requirement? What do you mean by exact match?
Query: term1 term2
Doc1: term1 term2
Doc2: term1 term2 term3 term4
Here both doc1 and doc2 are "exact match" or just Doc1?
If you want to capture the "how many query terms matched" information may be you can use http://lucene.apache.org/java/2_4_0/api/org/apache/lucene/search/Explanation.html
Or you can modify similarity to ignore all things other than coord, so that your documents are sorted by "how many of the query terms are found in the specified document"
http://lucene.apache.org/java/2_4_0/api/org/apache/lucene/search/Similarity.html#formula_coord
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org
Re: Search Score percentage, Should not be relative to the highest
score
Posted by Amr ElAdawy <Am...@etisalat.com>.
Hi iorixxx, Thanks a lot for your reply
I had read the link and I understand the concern, however, the normalization
is happening inside lucene. Where the normalizing value is the inverse of
the maxScore.
I can alter the code to leave the original score, however it is a business
requirements to view the matching percentage. Also, the absolute score means
nothing because we wont know the exact matching score.
I am thinking of alter the core to make the normalization relative to the
number of terms, considering that the each term will score 1, so the exact
match should score 3 if there are 3 terms.
The problem for that, I dont know how to get the number of Clauses from
Query, nor the number of Terms.
Also, some exact matching results scored more than 3 !!
any ideas?
--
View this message in context: http://lucene.472066.n3.nabble.com/Search-Score-percentage-Should-not-be-relative-to-the-highest-score-tp2183420p2183784.html
Sent from the Lucene - Java Users mailing list archive at Nabble.com.
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org
Re: Search Score percentage, Should not be relative to the highest score
Posted by Ahmet Arslan <io...@yahoo.com>.
It is somehow not recommended to convert scores to percentages.
http://wiki.apache.org/lucene-java/ScoresAsPercentages
> When using lucene to search documents, the results have a
> score based on their relativity to the search term. Inside
> lucene, the score
> percentage is calculated as a percentage of the maximum
> score achieved.
>
> Assume we are searching for Amr ElAdawy
>
> We get results like the following along with the score:
> Omar ElAdawy --> 1.6
> Amro ElAdawy --> 1.9
> Amir Adawi --> 1.3
>
>
> With no exact match. Then the percentage will be calculated
> relative to the
> highest score ( despite the fact that it is not exact
> match). So the
> percentage will be:
>
> Omar ElAdawy --> 84.21%
> Amro ElAdawy --> 100%
> Amir Adawi --> 68.42%
>
>
> I need to change that, so the percentage will be relative
> to the exact match
> score. And will be something like that (assuming the
> Matching result score
> is 3):
>
> Omar ElAdawy --> 53.33%
> Amro ElAdawy --> 63.33%
> Amir Adawi --> 43.33%
>
>
> using lucene 2.9.4
>
> Regards,
> Amr ElAdawy
>
>
>
>
> --
> View this message in context: http://lucene.472066.n3.nabble.com/Search-Score-percentage-Should-not-be-relative-to-the-highest-score-tp2183420p2183420.html
> Sent from the Lucene - Java Users mailing list archive at
> Nabble.com.
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-user-help@lucene.apache.org
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org