[jira] [Updated] (LUCENE-5252) add NGramSynonymTokenizer

["Koji Sekiguchi (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/LUCENE-5252?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Koji Sekiguchi updated LUCENE-5252:
-----------------------------------

    Attachment: LUCENE-5252_4x.patch

The draft patch without tests.

When NGramSynonymTokenizer was developed in RONDHUIT, it used double array trie for the synonym dictionary.

I've tried to convert the code to Lucene's FST. As this is the first experience of FST for me, any inefficient code may be there. Comments are welcome!

> add NGramSynonymTokenizer
> -------------------------
>
>                 Key: LUCENE-5252
>                 URL: https://issues.apache.org/jira/browse/LUCENE-5252
>             Project: Lucene - Core
>          Issue Type: New Feature
>          Components: modules/analysis
>            Reporter: Koji Sekiguchi
>            Priority: Minor
>         Attachments: LUCENE-5252_4x.patch
>
>
> I'd like to propose that we have another n-gram tokenizer which can process synonyms. That is NGramSynonymTokenizer. Note that in this ticket, the gram size is fixed, i.e. minGramSize = maxGramSize.
> Today, I think we have the following problems when using SynonymFilter with NGramTokenizer. 
> For purpose of illustration, we have a synonym setting "ABC, DEFG" w/ expand=true and N = 2 (2-gram).
> # There is no consensus (I think :-) how we assign offsets to generated synonym tokens DE, EF and FG when expanding source token AB and BC.
> # If the query pattern looks like XABC or ABCY, it cannot be matched even if there is a document "…XABCY…" in index when autoGeneratePhraseQueries set to true, because there is no "XA" or "CY" tokens in the index.
> NGramSynonymTokenizer can solve these problems by providing the following methods.
> * NGramSynonymTokenizer reads synonym settings (synonyms.txt) and it doesn't tokenize registered words. e.g.
> ||source text||NGramTokenizer+SynonymFilter||NGramSynonymTokenizer||
> |ABC|AB/DE/BC/EF/FG|ABC/DEFG|
> * The back and forth of the registered words, NGramSynonymTokenizer generates *extra* tokens w/ posInc=0. e.g.
> ||source text||NGramTokenizer+SynonymFilter||NGramSynonymTokenizer||
> |XYZABC123|XY/YZ/ZA/AB/DE/BC/EF/C1/FG/12/23|XY/YZ/Z/ABC/DEFG/1/12/23|
> In the above sample, "Z" and "1" are the extra tokens.



--
This message was sent by Atlassian JIRA
(v6.1#6144)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org