Multi match query and the scoring calculation in Elasticsearch - elasticsearch

Question

I have couple documents in the index, and 2 of them are
{
"id" : "c0706549-d06c-4043-8086-1b4b3ec1ef95",
"title" : "Google Pixel XL",
"memory" : "4GB",
"quantity" : 3
}
{
"id" : "23ecaecd-6b3f-4592-b79f-f46a20157221",
"title" : "Google Pixel XL",
"memory" : "6GB",
"quantity" : 1
}
And for the query
{
"query": { "multi_match": { "query": "pixel xl 6gb", "fields": ["title", "memory"] } }
}
I get the response
{
"_index" : "my_index",
"_type" : "_doc",
"_id" : "c0706549-d06c-4043-8086-1b4b3ec1ef95",
"_score" : 2.4280763,
"_source" : {
"id" : "c0706549-d06c-4043-8086-1b4b3ec1ef95",
"title" : "Google Pixel XL",
"memory" : "4GB",
"quantity" : 3
}
},
{
"_index" : "my_index",
"_type" : "_doc",
"_id" : "23ecaecd-6b3f-4592-b79f-f46a20157221",
"_score" : 2.4280763,
"_source" : {
"id" : "23ecaecd-6b3f-4592-b79f-f46a20157221",
"title" : "Google Pixel XL",
"memory" : "6GB",
"quantity" : 1
}
}
But I expect that the document with the memory field 6GB will be on top, can you please advise why this happens and how to fix it?
Index mapping
{
"mappings" : {
"properties" : {
"memory" : {
"type" : "text",
"fields" : {
"keyword" : {
"type" : "keyword",
"ignore_above" : 256
}
},
"fielddata" : true
},
"title" : {
"type" : "text",
"analyzer" : "synonym_analyzer"
}
}
}
}
Index settings
{
"index" : {
"analysis" : {
"filter" : {
"synonym_filter" : {
"type" : "synonym",
"synonyms" : [
"laptop, notebook"
]
}
},
"analyzer" : {
"synonym_analyzer" : {
"tokenizer" : "standard",
"filter" : ["lowercase", "synonym_filter"]
}
}
}
}
}
Elasticsearch version 7.7.0

Answer 1

I just tried this locally and I am getting much higher score(~2X) for a document containing the 6GB and if you note carefully in your case, both the document has exactly the same score (2.4280763), that means both of them has exactly the same relevance and its just the order in Elasticsearch response that is different.
Think of it, you need to sort the numbers 1,2,3,1 then 1,1,2,3 will be the order
it doesn't matter which 1 comes before or after.
Also, you need to provide your mapping and index configuration(number of shards) and elasticsearch version(as older version uses tf/idf while the new one uses BM25) for score calculation.
I tried this on ES 7.7 version and as mentioned earlier with my mapping and your sample data got 2X better score.
Index mapping
{
"mappings": {
"properties": {
"title" : {
"type": "text"
},
"memory" : {
"type" : "text"
}
}
}
}
Index your 2 docs
{
"title" : "Google Pixel XL",
"memory" : "6GB"
}
{
"title" : "Google Pixel XL",
"memory" : "4GB"
}
Search query
{
"query": {
"multi_match": {
"query": "pixel xl 6gb",
"fields": [
"title",
"memory"
]
}
}
}
And search result
"hits": [
{
"_index": "multma",
"_type": "_doc",
"_id": "2",
"_score": 0.6931471, --> note this
"_source": {
"title": "Google Pixel XL",
"memory": "6GB" --> 6 GB one has a better score and coming on top
}
},
{
"_index": "multma",
"_type": "_doc",
"_id": "1",
"_score": 0.36464313,
"_source": {
"title": "Google Pixel XL",
"memory": "4GB"
}
}
]