Implementiing "Partial match anywhere"

[David_Borsodi]

Hi

I'm trying to do a partial match anywhere search over medical terminologies. I've created any index with a multi-filed mapping of ngrams AND standard tokens.
Unfortunately I can't get a good relevance score in the case where a full word and an fragment is typed.

It seems that that full word+segment with AND operator is not a match on the standard index, but without the AND operator false positives come up (but otherwise the result ordering is pretty ok). If I use the AND operator, the standard/english query returns with a 0 and while there are no false positives, matching only one of the terms, the ordering becomes messed up.
This is my latest try:

    "query": {
    "dis_max": {
        "queries": [{
            "multi_match": {
                "query": expr,
                "operator": "AND",
                "type": "most_fields",
                "fields": [field, field + ".std^4"]
            }
        },
            {
                "multi_match": {
                    "query": expr,
                    "type": "most_fields",
                    "fields": [field + ".std^4"]
                }
            }
        ]
    }

Is there a standard way to solve this issue? (I forgot to mention that I'm new to ES and only read half of the definitive guide so far)

[David_Borsodi]

Experimenting further, it seems that splitting the expression to words and creating a bool-should for each of them increases the quality of the search results a lot. The false positives still show up though, but with lower relevance.
Nonetheless I'd be interested in any advice to filter them out completely
> {

            "bool": {
                "should": [
                    {
                        "term": {
                            "en.std": "neurological"
                        }
                    },
                    {
                        "term": {
                            "en.std": "diso"
                        }
                    }
                ]
            }
        }