[
https://issues.apache.org/jira/browse/LUCENE-7823?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16006515#comment-16006515
]
Tommaso Teofili commented on LUCENE-7823:
-----------------------------------------
checked in {{BM25NBClassifier}} implementation; when compared with
{{SimpleNaiveBayesClassifier}}, it gives a 0.06 improvement in f1 over 20
newsgroups dataset.
> Have a naive bayes classifier which uses plain BM25 scores instead of plain
> frequencies
> ---------------------------------------------------------------------------------------
>
> Key: LUCENE-7823
> URL: https://issues.apache.org/jira/browse/LUCENE-7823
> Project: Lucene - Core
> Issue Type: Improvement
> Components: modules/classification
> Reporter: Tommaso Teofili
> Assignee: Tommaso Teofili
> Fix For: master (7.0)
>
>
> {{SimpleNaiveBayesClassifier}} users term frequencies with add one smoothing
> to calculate likelihood and just tf for prior. Given Lucene has switched to
> BM25 it would be better to have a different impl which uses BM25
> scoring as a probability measure of both prior and likelihood.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
