Modifying elasticsearch score based on nested field value - search

Question

I want to modify scoring in ElasticSearch (v2+) based on the weight of a field in a nested object within an array.
For instance, using this data:
PUT index/test/0
{
"name": "red bell pepper",
"words": [
{"text": "pepper", "weight": 20},
{"text": "bell","weight": 10},
{"text": "red","weight": 5}
]
}
PUT index/test/1
{
"name": "hot red pepper",
"words": [
{"text": "pepper", "weight": 15},
{"text": "hot","weight": 11},
{"text": "red","weight": 5}
]
}
I want a query like {"words.text": "red pepper"} which would rank "red bell pepper" above "hot red pepper".
The way I am thinking about this problem is "first match the 'text' field, then modify scoring based on the 'weight' field". Unfortunately I don't know how to achieve this, if it's even possible, or if I have the right approach for something like this.
If proposing alternative approach, please try and keep a generalized idea where there are tons of different similar cases (eg: simply modifying the "red bell pepper" document score to be higher isn't really a suitable alternative).

Answer 1

The approach you have in mind is feasible. It can be achieved via function score in a nested query .
An example implementation is shown below :
PUT test
PUT test/test/_mapping
{
"properties": {
"name": {
"type": "string"
},
"words": {
"type": "nested",
"properties": {
"text": {
"type": "string"
},
"weight": {
"type": "long"
}
}
}
}
}
PUT test/test/0
{
"name": "red bell pepper",
"words": [
{"text": "pepper", "weight": 20},
{"text": "bell","weight": 10},
{"text": "red","weight": 5}
]
}
PUT test/test/1
{
"name": "hot red pepper",
"words": [
{"text": "pepper", "weight": 15},
{"text": "hot","weight": 11},
{"text": "red","weight": 5}
]
}
post test/_search
{
"query": {
"bool": {
"disable_coord": true,
"must": [
{
"match": {
"name": "red pepper"
}
}
],
"should": [
{
"nested": {
"path": "words",
"query": {
"function_score": {
"functions": [
{
"field_value_factor": {
"field" : "words.weight",
"missing": 0
}
}
],
"query": {
"match": {
"words.text": "red pepper"
}
},
"score_mode": "sum",
"boost_mode": "replace"
}
},
"score_mode": "total"
}
}
]
}
}
}
Result :
"hits": [
{
"_index": "test",
"_type": "test",
"_id": "0",
"_score": 26.030865,
"_source": {
"name": "red bell pepper",
"words": [
{
"text": "pepper",
"weight": 20
},
{
"text": "bell",
"weight": 10
},
{
"text": "red",
"weight": 5
}
]
}
},
{
"_index": "test",
"_type": "test",
"_id": "1",
"_score": 21.030865,
"_source": {
"name": "hot red pepper",
"words": [
{
"text": "pepper",
"weight": 15
},
{
"text": "hot",
"weight": 11
},
{
"text": "red",
"weight": 5
}
]
}
}
]
}
The query in a nutshell would score a document that satisfies the must clause as follows : sum up the weights of the matched nested documents with the score of the must clause.