Elastic Search - Even distribution on a map - google-maps

Question

I'm using Elastic Search 2. I have a big database of locations, all of them have a gps attribute, which is a geopoint.
My frontend application displays a google maps component with the results, filtered by my query, let's say pizza. The problem is that the dataset grew a lot, and the client wants even results on the map.
So if I search for a specific query in New York, i would like to have results all over New York, but i'm currently receiving 400 results only in one populous area of Manhattan.
My naive approach was to just filter by distance
{
"size":400,
"query":{
"bool":{
"must":{
"match_all":{
}
},
"filter":{
"geo_distance":{
"distance":"200km",
"gps":[
-73.98502023369585,
40.76195656809083
]
}
}
}
}
}
This doesn't guarantee that the results will be spread across the map.
How can I do it?
I've tried using Geo-Distance Aggregation for this
{
"size":400,
"query":{
"bool":{
"must":{
"match_all":{
}
},
"filter":{
"geo_distance":{
"distance":"200km",
"gps":[
-73.98502023369585,
40.76195656809083
]
}
}
}
},
"aggs":{
"per_ring":{
"geo_distance":{
"field":"gps",
"unit":"km",
"origin":[
-73.98502023369585,
40.76195656809083
],
"ranges":[
{
"from":0,
"to":100
},
{
"from":100,
"to":200
}
]
}
}
}
}
But i just receive a results list + the amount of elements that belong to the buckets. The results list is not guaranteed to be spread.
"aggregations": {
"per_ring": {
"buckets": [
{
"key": "*-100.0",
"from": 0,
"from_as_string": "0.0",
"to": 100,
"to_as_string": "100.0",
"doc_count": 33821
},
{
"key": "100.0-200.0",
"from": 100,
"from_as_string": "100.0",
"to": 200,
"to_as_string": "200.0",
"doc_count": 6213
}
]
}
}
I would like to grab half of the results from one bucket, half from the other bucket.
I've also attempted to use Geohash Grid Aggregation, but that also doesn't give me samples of results for every bucket, just provides the areas.
So how do I get a spaced distribution of results spread across my map with one elastic search query?
Thanks!

Answer 1

I think introducing some randomness may give you the desired result. I am assuming you're seeing the same distribution because of index ordering (you're not scoring based on distance, and you're taking the first 400 so you are most likely seeing the same result set).
{
"size": 400,
"query": {
"function_score": {
"query": {
"bool": {
"must": [
{
"match_all": {}
}
],
"filter": {
"geo_distance": {
"distance": "200km",
"gps": [
-73.98502023369585,
40.76195656809083
]
}
}
}
},
"functions": [
{
"random_score": {}
}
]
}
}
}
Random score in elastic