Elastic Bytes - rank/distance feature

examples.json

###################################################
## Boosting scores based on numeric field values ##
###################################################

# Existing solution: function_score query
# Problem: Performance penalty
# Solution: Rescoring, only rescoring top-n documents

# Lucene optimization in Elasticsearch 7: Block MAX WAND
# New query: distance_feature, efficiently skips non-competitive hits
# Works with geopoint, date and date_nanos

GET /

PUT hotels 
{
  "mappings": {
    "properties": {
      "location" : { "type" : "geo_point" },
      "last_booked" : { "type" : "date" }
    }
  }
}

PUT hotels/_doc/motel-one-central
{
  "name" : "Motel One - Munich Sonnenstraße",
  "last_booked" : "2021-03-02",
  "stars": 5,
  "location": { "lat" : 48.1345, "lon":11.5669 }
}

PUT hotels/_doc/motel-one-schwabing
{
  "name" : "Motel One - Munich Parkstadt Schwabing",
  "last_booked" : "2021-01-01",
  "stars": 4,
  "location": { "lat" : 48.1778 , "lon":11.5919 }
}

PUT hotels/_doc/ibis-west?refresh
{
  "name" : "Ibis Hotel/Motel - München Süd/West",
  "last_booked" : "2020-12-15",
  "stars": 2,
  "location": { "lat" : 48.1347 , "lon":11.5241 }
}


GET hotels/_search

GET hotels/_search
{
  "query": {
    "bool": {
      "must": [
        {
          "match": {
            "name": "motel"
          }
        }
      ]
    }
  }
}

# score low budget hotels higher
GET hotels/_search
{
  "query": {
    "bool": {
      "must": [
        {
          "match": {
            "name": "motel"
          }
        }
      ],
      "should": [
        {
          "range": {
            "stars": {
              "lte": 3
            }
          }
        }
      ]
    }
  }
}

# main station: 48.14, 11.5595
# let's give those hotels a boost next to the main station
GET hotels/_search
{
  "query": {
    "bool": {
      "must": [
        {
          "match": {
            "name": "motel"
          }
        }
      ],
      "should": [
        {
          "distance_feature": {
            "field": "location",
            "origin": { "lat" : 48.14, "lon" : 11.5595},
            "pivot": "1km"
          }
        }
      ]
    }
  }
}

# BMW World/Olympiapark: 48.1776, 11.5561
GET hotels/_search
{
  "query": {
    "bool": {
      "must": [
        {
          "match": {
            "name": "motel"
          }
        }
      ],
      "should": [
        {
          "distance_feature": {
            "field": "location",
            "origin": { "lat" : 48.1776, "lon" : 11.5561},
            "pivot": "3km"
          }
        }
      ]
    }
  }
}

# airport location: 48.3539, 11.7882
GET hotels/_search
{
  "query": {
    "bool": {
      "must": [
        {
          "match": {
            "name": "motel"
          }
        }
      ],
      "should": [
        {
          "distance_feature": {
            "field": "location",
            "origin": { "lat" : 48.3539, "lon" : 11.7882},
            "pivot": "1km"
          }
        }
      ]
    }
  }
}


# score hotels higher that have been booked more recently
GET hotels/_search
{
  "query": {
    "bool": {
      "must": [
        {
          "range": {
            "stars": {
              "gte": 2
            }
          }
        }
      ],
      "should": [
        {
          "distance_feature": {
            "field": "last_booked",
            "origin": "2021-03-17",
            "pivot": "30d"
          }
        }
      ]
    }
  }
}

# but how to boost by four star rating?
# well distance feature only supports geo points and dates
# Alternative: rank_feature
POST hotels/_mapping
{
  "properties" : {
    "rating" : {
      "type" : "rank_feature"
    }
  }
}

GET hotels/_mapping

POST hotels/_update/motel-one-central
{
  "doc" : {
    "rating" : 5.0
  }
}


POST hotels/_update/motel-one-schwabing
{
  "doc" : {
    "rating" : 3.8
  }
}

POST hotels/_update/ibis-west?refresh
{
  "doc" : {
    "rating" : 3.1,
    "name" : "Ibis Motel - München West"
  }
}

# First two same score
GET hotels/_search
{
  "query": {
    "bool": {
      "must": [
        {
          "match": {
            "name": "motel"
          }
        }
      ]
    }
  }
}


# add explanation
GET hotels/_search
{
  "explain": true, 
  "query": {
    "bool": {
      "must": [
        {
          "match": {
            "name": "motel"
          }
        }
      ],
      "should": [
        {
          "rank_feature": {
            "field": "rating"
          }
        }
      ]
    }
  }
}

# difference rank_feature vs. distance features
# point of reference from the outside vs. document values

# both skip non competitive hits (biggest difference to function score)

# More info: https://www.elastic.co/guide/en/elasticsearch/reference/7.11/query-dsl-distance-feature-query.html
# More info: https://www.elastic.co/guide/en/elasticsearch/reference/7.11/query-dsl-rank-feature-query.html
# https://www.elastic.co/blog/easier-relevance-tuning-elasticsearch-7-0
# https://www.elastic.co/blog/distance-feature-query-time-and-geo-in-elasticsearch-result-ranking
# https://www.elastic.co/blog/faster-retrieval-of-top-hits-in-elasticsearch-with-block-max-wand

# Kahoot: https://kahoot.it/challenge/09692849