lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Arcadius Ahouansou (JIRA)" <>
Subject [jira] [Commented] (LUCENE-6939) BlendedInfixSuggester to support exponential reciprocal BlenderType
Date Sat, 14 May 2016 05:47:12 GMT


Arcadius Ahouansou commented on LUCENE-6939:

Hello  [~erickerickson]

I can see you have updated the wiki with {{linear}}->{{position_linear}} and {{reciprocal}}->{{position_reciprocal}}

Thank you very much for that  [~erickerickson] 

We need a new bullet point for the new blenderType, just below {{position_reciprocal}} at
the same level: 

{{position_exponential_reciprocal}}:  weightFieldValue/Math.pow(1+position, exponent). This
is a more aggressive and more configurable version of the {{position_reciprocal}} with a configurable
variable {{exponent}}. When {{exponent}}==1.0, then both {{position_exponential_reciprocal}}
and {{position_reciprocal}} are equivalent and lead to the same result. 


Then, we will need a new bullet point outside of the BlenderType and just above numFactor:
{{exponent}}: an optional configuration variable for the {{position_exponential_reciprocal}}
blenderType. It is a decimal number used to control how fast the score will grow or decrease.
Its default value 2.0

Many thanks

> BlendedInfixSuggester to support exponential reciprocal BlenderType
> -------------------------------------------------------------------
>                 Key: LUCENE-6939
>                 URL:
>             Project: Lucene - Core
>          Issue Type: Improvement
>          Components: modules/spellchecker
>    Affects Versions: 5.4
>            Reporter: Arcadius Ahouansou
>            Priority: Minor
>              Labels: suggester
>             Fix For: 5.5, 6.0
>         Attachments: LUCENE-6939.patch
> The orignal BlendedInfixSuggester introduced in LUCENE-5354 has support for:
> - {{BlenderType.POSITION_LINEAR}} and 
> - {{BlenderType.POSITION_RECIPROCAL}} .
> These are used to score documents based on the position of the matched token i.e the
closer is the matched term to the beginning, the higher score you get.
> In some use cases, we need a more aggressive scoring based on the position.
> That's where the exponential reciprocal comes into play 
> i.e 
> {{coef = 1/Math.pow(position+1, exponent)}}
> where the {{exponent}} is a configurable variable.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message