Sorting by distance with minimum distance

Elastic Stack Elasticsearch
1

I'm trying to sort results by distance. Something like this works perfectly:

_geo_distance: {
  "nested_path": ...,
  "location": [..., ...],
  "order": "asc",
  "unit": "m",
  "mode": "min",
  "distance_type": "sloppy_arc"
}

However I would like to use an minimum distance (if under minimum distance all elements should have the same weight in result). So I've tried

_geo_distance: {
  "nested_path": ...,
  "location": {
    "type": "circle",
    "coordinates": [..., ...],
    "radius": "100m"
  }
  "order": "asc",
  "unit": "m",
  "mode": "min",
  "distance_type": "sloppy_arc"
}

But that doesn't seem to work. Is there an easy way to accomplish what I want to do? Do I have to go with scoring?

2

Anyone has an idea?

3

ping

4

I would go with scoring. The function_score query supports decay functions which will allow you to score documents based on a geo distance. These decay functions allow you to define an offset, which would be a radius for which documents get a maximum score. Any document with a distance larger than this offset will get a lower score, based on how you configure the decay function.

The docs have an example: https://www.elastic.co/guide/en/elasticsearch/reference/current/query-dsl-function-score-query.html#function-decay

5

Sweet, thank you very much for your reply! I'll hopefully get to implementing this in the next few weeks and will try to post our final solution here.

6

Something like this seems to get the job done:

      "functions": [
        {
          "exp": {
            "venues.location": {
              "origin": [
                0.5, 
                0.5
              ], 
              "decay": "0.5", 
              "scale": "20km", 
              "offset": "100m"
            }
          }
        },
        ...
      ]

This works for nested objects as well if you use include_in_root on the nesting.

7

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.