[ 
https://issues.apache.org/jira/browse/LUCENE-5350?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13837405#comment-13837405
 ] 

Areek Zillur commented on LUCENE-5350:
--------------------------------------

Thanks for the feedback!
[~mikemccand]: I was wondering the same thing as I was implementing it. By 
intuition I think this should be more compact than the N suggester approach but 
I think its best to benchmark it against N separate suggesters (will update 
with the benchmark results). Also had another idea of implementing this by 
'filtering' the suggestions by contexts supplied at query-time rather than 
prefixing the context with the analyzed form. I will play around with both and 
benchmark it to see whether this would be useful in practice.
[~ppujari]: Can you expand on the deep learning algorithm part? 

> Add Context Aware Suggester
> ---------------------------
>
>                 Key: LUCENE-5350
>                 URL: https://issues.apache.org/jira/browse/LUCENE-5350
>             Project: Lucene - Core
>          Issue Type: New Feature
>          Components: core/search
>            Reporter: Areek Zillur
>             Fix For: 5.0, 4.7
>
>         Attachments: LUCENE-5350.patch
>
>
> It would be nice to have a Context Aware Suggester (i.e. a suggester that 
> could return suggestions depending on some specified context(s)).
> Use-cases: 
>   - location-based suggestions:
>       -- returns suggestions which 'match' the context of a particular area
>           --- suggest restaurants names which are in Palo Alto (context -> 
> Palo Alto)
>   - category-based suggestions:
>       -- returns suggestions for items that are only in certain 
> categories/genres (contexts)
>           --- suggest movies that are of the genre sci-fi and adventure 
> (context -> [sci-fi, adventure])



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

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to