Hi,
I've spent several days trying to achieve something like this in ES:
http://www.lucidimagination.com/search/link?url=http://wiki.apache.org/solr/Payloads
But had no luck.
This is what I did (maybe all of that in unnecessary, but I couldn't find
any information on how to do this correctly):
-
Created a custom similarity and a provider (mimicking
https://github.com/tlrx/elasticsearch-custom-similarity-provider )@Override
public float scorePayload(int docId, java.lang.String fieldName, int
start, int end, byte[] payload, int offset, int length) {if (payload != null) { return PayloadHelper.decodeFloat(payload, offset); } else { return 1.0F; }}
-
Created a custom analyzer:
"analyzer": { "pf": { "tokenizer": "whitespace", "filter": "standard, lowercase, stop, payload" } }
where the payload filter just does:
@Override
public TokenStream create(TokenStream tokenStream) {
return new DelimitedPayloadTokenFilter(tokenStream, '|', new
FloatEncoder());
}
- created an index that uses that similarity for index and search and that
analyzer. - Created a mapping that instructs to use the new analyzer when indexing
field "message" - indexed a document:
curl -XPUT 'http://localhost:9200/twitter/tweet/1' -d '{
"tweet" : {
"user" : "kimchy",
"message" : "search|40.0"
}
}'
Did a search for term "search", but looking at the query "explain" the
payload doesn't seem to be taken into account.
Also when I retrieve the document back, I see the content of "message" to
be exactly like I submitted to indexing (including the pipe and payload)
not sure this is a bad sign.
Please let me know what I'm doing wrong here.