Elastic Search: Any way to make space-separated words in a comma-separated list regarded as one term? - match

Question

I don't know if this is possible, but I'm trying to search by locations with an "exact search" option. There are a couple fields that get searched, with the most important one being the "location_raw" field:
"match": {
"location.location_raw": {
"type": "boolean",
"operator": "AND",
"query": "[location query]",
"analyzer": "standard"
}
}
The location_raw field is a location string with a comma between each place, such as "Sudbury, Middlesex, Massachusetts" or "Leamington, Warwickshire, England". If someone searches for "Sudbury, Middlesex" it gets passed in as
"query": "Sudbury Middlesex"
and both of those terms must exist in the location_raw field. This part works.
The problem is that when the location_raw field contains multi-word location, like New York or Saint George, these get returned when someone searches for "York" or "George." If I do an exact search for "George," I do not want to get results for "Saint George." Is there any way to make Elastic consider "Saint George" one term in the string "Saint George, Stamford, Lincoln, England"?

Answer 1

Here's one way to do it, but you have to query in csv too, or use a terms filter.
I used a pattern analyzer with a simple pattern: ", ". I set up a simple index with a single document:
PUT /test_index
{
"settings": {
"number_of_shards": 1,
"analysis": {
"analyzer": {
"csv": {
"type": "pattern",
"pattern": ", ",
"lowercase": false
}
}
}
},
"mappings": {
"doc": {
"properties": {
"location": {
"type": "string",
"index_analyzer": "csv",
"search_analyzer": "standard",
"fields": {
"raw": {
"type": "string",
"index": "not_analyzed"
}
}
}
}
}
}
}
POST /test_index/_bulk
{"index":{"_index":"test_index","_type":"doc","_id":1}}
{"location":"Saint George, Stamford, Lincoln, England"}
I can see the terms generated with a simple terms aggregation:
POST /test_index/_search?search_type=count
{
"aggs": {
"location_terms": {
"terms": {
"field": "location"
}
}
}
}
...
{
"took": 1,
"timed_out": false,
"_shards": {
"total": 1,
"successful": 1,
"failed": 0
},
"hits": {
"total": 1,
"max_score": 0,
"hits": []
},
"aggregations": {
"location_terms": {
"buckets": [
{
"key": "England",
"doc_count": 1
},
{
"key": "Lincoln",
"doc_count": 1
},
{
"key": "Saint George",
"doc_count": 1
},
{
"key": "Stamford",
"doc_count": 1
}
]
}
}
}
And then if I query with the same csv syntax, the document isn't returned for "George, England":
POST /test_index/_search
{
"query": {
"match": {
"location": {
"type": "boolean",
"operator": "AND",
"query": "George, England",
"analyzer": "csv"
}
}
}
}
...
{
"took": 0,
"timed_out": false,
"_shards": {
"total": 1,
"successful": 1,
"failed": 0
},
"hits": {
"total": 0,
"max_score": null,
"hits": []
}
}
but is for "Saint George, England":
POST /test_index/_search
{
"query": {
"match": {
"location": {
"type": "boolean",
"operator": "AND",
"query": "Saint George, England",
"analyzer": "csv"
}
}
}
}
...
{
"took": 1,
"timed_out": false,
"_shards": {
"total": 1,
"successful": 1,
"failed": 0
},
"hits": {
"total": 1,
"max_score": 0.2169777,
"hits": [
{
"_index": "test_index",
"_type": "doc",
"_id": "1",
"_score": 0.2169777,
"_source": {
"location": "Saint George, Stamford, Lincoln, England"
}
}
]
}
}
This query is equivalent, and probably more performant:
POST /test_index/_search
{
"query": {
"filtered": {
"query": {
"match_all": {}
},
"filter": {
"terms": {
"location": [
"Saint George",
"England"
],
"execution": "and"
}
}
}
}
}
Here's the code I used to test it:
http://sense.qbox.io/gist/234ea93accb7b20ad8fd33e62fe92f1d450a51ab