Elasticsearch: Why is score different from value in explanation?
Hi,
I did an explain request and get the following response:
"_score": 2,
"fields": {
"title": [
"Title"
],
"id": [
"vbce77e7t9pe"
]
},
"highlight": {
"content.search": [
"... some content ..."
]
},
"_explanation": {
"value": 196.23717,
"description": "sum of:",
"details": [
{...
},
{
"value": 4.737492,
"description": "product of:",
"details": [...
]
}
]
},
As you can see the score value is 2 and the value in the explanation is 196.23717.
Why is this different?
krueger
All 9 comments
This is a bug, they should always be the same. Can you share your query?
cc @elastic/es-search-aggs
jpountz
on 19 Feb 2018
This is the query
{
"from" : 0,
"size" : 67,
"query" : {
"function_score" : {
"query" : {
"bool" : {
"must" : [
{
"bool" : {
"should" : [
{
"nested" : {
"query" : {
"bool" : {
"must" : [
{
"constant_score" : {
"filter" : {
"bool" : {
"must" : [
{
"terms" : {
"elements.aclgroups" : [
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"DUMMY_GROUP_ID"
],
"boost" : 1.0
}
},
{
"term" : {
"elements.hidden" : {
"value" : false,
"boost" : 1.0
}
}
},
{
"bool" : {
"should" : [
{
"term" : {
"elements.approval_status" : {
"value" : "APPROVED",
"boost" : 1.0
}
}
},
{
"bool" : {
"must_not" : [
{
"exists" : {
"field" : "elements.approval_status",
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
},
{
"bool" : {
"must" : [
{
"range" : {
"elements.valid_from" : {
"from" : null,
"to" : "now",
"include_lower" : true,
"include_upper" : true,
"time_zone" : "+01:00",
"boost" : 1.0
}
}
},
{
"range" : {
"elements.valid_to" : {
"from" : "now",
"to" : null,
"include_lower" : true,
"include_upper" : true,
"time_zone" : "+01:00",
"boost" : 1.0
}
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
},
"boost" : 1.0
}
},
{
"bool" : {
"must" : [
{
"multi_match" : {
"query" : "<query>",
"fields" : [
"elements.content.search^1.0",
"elements.tags^1.0",
"elements.title.search^1.0"
],
"type" : "best_fields",
"operator" : "AND",
"slop" : 0,
"prefix_length" : 0,
"max_expansions" : 50,
"lenient" : false,
"cutoff_frequency" : 0.0,
"zero_terms_query" : "NONE",
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
},
"path" : "elements",
"ignore_unmapped" : true,
"score_mode" : "none",
"boost" : 1.0,
"inner_hits" : {
"name" : "elements",
"ignore_unmapped" : true,
"from" : 0,
"size" : 3,
"version" : false,
"explain" : false,
"track_scores" : false,
"highlight" : {
"pre_tags" : [
"<span class=\"highlighted\">"
],
"post_tags" : [
"</span>"
],
"fragment_size" : 80,
"number_of_fragments" : 3,
"type" : "fvh",
"fields" : {
"elements.content" : {
"type" : "fvh",
"matched_fields" : [
"elements.content",
"elements.content.search"
]
}
}
}
}
}
},
{
"bool" : {
"must" : [
{
"constant_score" : {
"filter" : {
"bool" : {
"must" : [
{
"bool" : {
"should" : [
{
"term" : {
"aclusers" : {
"value" : "<some id>",
"boost" : 1.0
}
}
},
{
"bool" : {
"must" : [
{
"terms" : {
"resource" : [
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>"
],
"boost" : 1.0
}
},
{
"terms" : {
"aclwritegroups" : [
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"<some id>",
"DUMMY_GROUP_ID"
],
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
},
"boost" : 1.0
}
},
{
"nested" : {
"query" : {
"bool" : {
"must" : [
{
"multi_match" : {
"query" : "<query>",
"fields" : [
"elements.content.search^1.0",
"elements.tags^1.0",
"elements.title.search^1.0"
],
"type" : "best_fields",
"operator" : "AND",
"slop" : 0,
"prefix_length" : 0,
"max_expansions" : 50,
"lenient" : false,
"cutoff_frequency" : 0.0,
"zero_terms_query" : "NONE",
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
},
"path" : "elements",
"ignore_unmapped" : true,
"score_mode" : "none",
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
},
{
"bool" : {
"must" : [
{
"bool" : {
"must_not" : [
{
"terms" : {
"resource" : [
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>",
"<some resource>"
],
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
},
"functions" : [
{
"filter" : {
"match_all" : {
"boost" : 1.0
}
},
"field_value_factor" : {
"field" : "boost_factor",
"factor" : 1.0,
"modifier" : "none"
}
}
],
"score_mode" : "multiply",
"max_boost" : 3.4028235E38,
"boost" : 1.0
}
},
"explain" : false,
"stored_fields" : [
"id",
"title"
],
"highlight" : {
"pre_tags" : [
"<span class=\"highlighted\">"
],
"post_tags" : [
"</span>"
],
"type" : "fvh",
"fields" : {
"content.search" : {
"fragment_size" : 80,
"number_of_fragments" : 3,
"type" : "fvh",
"highlight_query" : {
"bool" : {
"must" : [
{
"multi_match" : {
"query" : "<query>",
"fields" : [
"content.search^1.0"
],
"type" : "best_fields",
"operator" : "AND",
"slop" : 0,
"prefix_length" : 0,
"max_expansions" : 50,
"lenient" : false,
"cutoff_frequency" : 0.0,
"zero_terms_query" : "NONE",
"boost" : 1.0
}
}
],
"disable_coord" : false,
"adjust_pure_negative" : true,
"boost" : 1.0
}
},
"matched_fields" : [
"content.search"
]
}
}
},
"suggest" : {
"autocomplete" : {
"text" : "<query>",
"term" : {
"field" : "autocomplete",
"suggest_mode" : "MISSING",
"accuracy" : 0.5,
"sort" : "SCORE",
"string_distance" : "INTERNAL",
"max_edits" : 2,
"max_inspections" : 5,
"max_term_freq" : 0.01,
"prefix_length" : 1,
"min_word_length" : 4,
"min_doc_freq" : 0.0
}
}
},
"rescore" : [
{
"query" : {
"rescore_query" : {
"multi_match" : {
"query" : "<query>",
"fields" : [
"elements.content.stem^1.0",
"elements.tags^1.0",
"elements.title.stem^1.0"
],
"type" : "phrase",
"operator" : "OR",
"slop" : 0,
"prefix_length" : 0,
"max_expansions" : 50,
"lenient" : false,
"zero_terms_query" : "NONE",
"boost" : 1.0
}
},
"query_weight" : 1.0,
"rescore_query_weight" : 80.0,
"score_mode" : "total"
}
},
{
"query" : {
"rescore_query" : {
"match" : {
"elements.content.stem" : {
"query" : "<query>",
"operator" : "AND",
"prefix_length" : 0,
"max_expansions" : 50,
"fuzzy_transpositions" : true,
"lenient" : false,
"zero_terms_query" : "NONE",
"boost" : 1.0
}
}
},
"query_weight" : 1.0,
"rescore_query_weight" : 2.0,
"score_mode" : "total"
}
}
]
}
Which value is used for sorting the search results? The value from _score or the value from _explanation? Is the result sorted correctly?
krueger
on 19 Feb 2018
Which value is used for sorting the search results? The value from _score or the value from _explanation? Is the result sorted correctly?
Results are always sorted based on _score, _explanation is just extra debugging information. I suspect results are still sorted correctly, but it is hard to say for sure without knowing what the exact bug is.
jpountz
on 19 Feb 2018
This is the complete explanation part:
"_explanation": {
"value": 196.23717,
"description": "sum of:",
"details": [
{
"value": 191.49968,
"description": "product of:",
"details": [
{
"value": 191.49968,
"description": "sum of:",
"details": [
{
"value": 2,
"description": "product of:",
"details": [
{
"value": 2,
"description": "sum of:",
"details": [
{
"value": 2,
"description": "function score, product of:",
"details": [
{
"value": 2,
"description": "sum of:",
"details": [
{
"value": 1,
"description": "sum of:",
"details": [
{
"value": 0,
"description": "Score based on 1 child docs in range from 2714 to 2714, best match:",
"details": [
{
"value": 7.6046343,
"description": "sum of:",
"details": [
{
"value": 7.6046343,
"description": "sum of:",
"details": [
{
"value": 1,
"description": "ConstantScore(+ConstantScore(elements.aclgroups:<id>
elements.aclgroups:<id> elements.aclgroups:<id> elements.aclgroups:<id> elements.aclgroups:<id> elements.aclgroups:<id> elements.aclgroups:<id> elements.aclgroups:<id> elements.aclgroups:<id> elements.aclgroups:<id> elements.aclgroups:<id> elements.aclgroups:DUMMY_GROUP_ID elements.aclgroups:ubguhplma88o) +elements.hidden:F +((elements.approval_status:APPROVED (-ConstantScore(_field_names:elements.approval_status) +*:*))~1) +(+elements.valid_from:[-9223372036854775808 TO 9223372036854775807] +elements.valid_to:[1519050328256 TO 9223372036854775807])), product of:",
"details": [
{
"value": 1,
"description": "boost",
"details": []
},
{
"value": 1,
"description": "queryNorm",
"details": []
}
]
},
{
"value": 6.6046343,
"description": "max of:",
"details": [
{
"value": 6.6046343,
"description": "weight(elements.content.search:term in 16) [PerFieldSimilarity], result of:",
"details": [
{
"value": 6.6046343,
"description": "score(doc=16,freq=1.0 = termFreq=1.0\n), product of:",
"details": [
{
"value": 3.9169664,
"description": "idf, computed as log(1 + (docCount - docFreq + 0.5) / (docFreq + 0.5)) from:",
"details": [
{
"value": 56,
"description": "docFreq",
"details": []
},
{
"value": 2838,
"description": "docCount",
"details": []
}
]
},
{
"value": 1.6861606,
"description": "tfNorm, computed as (freq * (k1 + 1)) / (freq + k1 * (1 - b + b * fieldLength / avgFieldLength)) from:",
"details": [
{
"value": 1,
"description": "termFreq=1.0",
"details": []
},
{
"value": 1.2,
"description": "parameter k1",
"details": []
},
{
"value": 0.75,
"description": "parameter b",
"details": []
},
{
"value": 1944.5743,
"description": "avgFieldLength",
"details": []
},
{
"value": 10.24,
"description": "fieldLength",
"details": []
}
]
}
]
}
]
}
]
}
]
},
{
"value": 0,
"description": "match on required clause, product of:",
"details": [
{
"value": 0,
"description": "# clause",
"details": []
},
{
"value": 1,
"description": "_type:__elements, product of:",
"details": [
{
"value": 1,
"description": "boost",
"details": []
},
{
"value": 1,
"description": "queryNorm",
"details": []
}
]
}
]
}
]
}
]
},
{
"value": 1,
"description": "sum of:",
"details": [
{
"value": 1,
"description": "ConstantScore((aclusers:4374ef463c8cb7f0013c967ea7cc4701 (+resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> resource:<resource> +ConstantScore(aclwritegroups:<id> aclwritegroups:<id> aclwritegroups:<id> aclwritegroups:<id> aclwritegroups:<id> aclwritegroups:<id> aclwritegroups:<id> aclwritegroups:<id> aclwritegroups:<id> aclwritegroups:<id> aclwritegroups:<id> aclwritegroups:DUMMY_GROUP_ID aclwritegroups:<id>)))~1), product of:",
"details": [
{
"value": 1,
"description": "boost",
"details": []
},
{
"value": 1,
"description": "queryNorm",
"details": []
}
]
},
{
"value": 0,
"description": "Score based on 1 child docs in range from 2714 to 2714, best match:",
"details": [
{
"value": 6.6046343,
"description": "sum of:",
"details": [
{
"value": 6.6046343,
"description": "max of:",
"details": [
{
"value": 6.6046343,
"description": "weight(elements.content.search:term in 16) [PerFieldSimilarity], result of:",
"details": [
{
"value": 6.6046343,
"description": "score(doc=16,freq=1.0 = termFreq=1.0\n), product of:",
"details": [
{
"value": 3.9169664,
"description": "idf, computed as log(1 + (docCount - docFreq + 0.5) / (docFreq + 0.5)) from:",
"details": [
{
"value": 56,
"description": "docFreq",
"details": []
},
{
"value": 2838,
"description": "docCount",
"details": []
}
]
},
{
"value": 1.6861606,
"description": "tfNorm, computed as (freq * (k1 + 1)) / (freq + k1 * (1 - b + b * fieldLength / avgFieldLength)) from:",
"details": [
{
"value": 1,
"description": "termFreq=1.0",
"details": []
},
{
"value": 1.2,
"description": "parameter k1",
"details": []
},
{
"value": 0.75,
"description": "parameter b",
"details": []
},
{
"value": 1944.5743,
"description": "avgFieldLength",
"details": []
},
{
"value": 10.24,
"description": "fieldLength",
"details": []
}
]
}
]
}
]
}
]
},
{
"value": 0,
"description": "match on required clause, product of:",
"details": [
{
"value": 0,
"description": "# clause",
"details": []
},
{
"value": 1,
"description": "_type:__elements, product of:",
"details": [
{
"value": 1,
"description": "boost",
"details": []
},
{
"value": 1,
"description": "queryNorm",
"details": []
}
]
}
]
}
]
}
]
}
]
}
]
},
{
"value": 1,
"description": "sum of:",
"details": [
{
"value": 1,
"description": "*:*, product of:",
"details": [
{
"value": 1,
"description": "boost",
"details": []
},
{
"value": 1,
"description": "queryNorm",
"details": []
}
]
}
]
}
]
},
{
"value": 1,
"description": "min of:",
"details": [
{
"value": 1,
"description": "field value function: none(doc['boost_factor'].value * factor=1.0)",
"details": []
},
{
"value": 3.4028235e+38,
"description": "maxBoost",
"details": []
}
]
}
]
},
{
"value": 0,
"description": "match on required clause, product of:",
"details": [
{
"value": 0,
"description": "# clause",
"details": []
},
{
"value": 1,
"description": "#*:* -_type:__*, product of:",
"details": [
{
"value": 1,
"description": "boost",
"details": []
},
{
"value": 1,
"description": "queryNorm",
"details": []
}
]
}
]
}
]
},
{
"value": 1,
"description": "primaryWeight",
"details": []
}
]
},
{
"value": 189.49968,
"description": "product of:",
"details": [
{
"value": 2.368746,
"description": "max of:",
"details": [
{
"value": 2.368746,
"description": "weight(elements.content.stem:term in 17) [PerFieldSimilarity], result of:",
"details": [
{
"value": 2.368746,
"description": "score(doc=17,freq=2.0 = termFreq=2.0\n), product of:",
"details": [
{
"value": 3.9530065,
"description": "idf, computed as log(1 + (docCount - docFreq + 0.5) / (docFreq + 0.5)) from:",
"details": [
{
"value": 54,
"description": "docFreq",
"details": []
},
{
"value": 2838,
"description": "docCount",
"details": []
}
]
},
{
"value": 0.5992265,
"description": "tfNorm, computed as (freq * (k1 + 1)) / (freq + k1 * (1 - b + b * fieldLength / avgFieldLength)) from:",
"details": [
{
"value": 2,
"description": "termFreq=2.0",
"details": []
},
{
"value": 1.2,
"description": "parameter k1",
"details": []
},
{
"value": 0.75,
"description": "parameter b",
"details": []
},
{
"value": 238.7012,
"description": "avgFieldLength",
"details": []
},
{
"value": 1337.4694,
"description": "fieldLength",
"details": []
}
]
}
]
}
]
}
]
},
{
"value": 80,
"description": "secondaryWeight",
"details": []
}
]
}
]
},
{
"value": 1,
"description": "primaryWeight",
"details": []
}
]
},
{
"value": 4.737492,
"description": "product of:",
"details": [
{
"value": 2.368746,
"description": "weight(elements.content.stem:term in 17) [PerFieldSimilarity], result of:",
"details": [
{
"value": 2.368746,
"description": "score(doc=17,freq=2.0 = termFreq=2.0\n), product of:",
"details": [
{
"value": 3.9530065,
"description": "idf, computed as log(1 + (docCount - docFreq + 0.5) / (docFreq + 0.5)) from:",
"details": [
{
"value": 54,
"description": "docFreq",
"details": []
},
{
"value": 2838,
"description": "docCount",
"details": []
}
]
},
{
"value": 0.5992265,
"description": "tfNorm, computed as (freq * (k1 + 1)) / (freq + k1 * (1 - b + b * fieldLength / avgFieldLength)) from:",
"details": [
{
"value": 2,
"description": "termFreq=2.0",
"details": []
},
{
"value": 1.2,
"description": "parameter k1",
"details": []
},
{
"value": 0.75,
"description": "parameter b",
"details": []
},
{
"value": 238.7012,
"description": "avgFieldLength",
"details": []
},
{
"value": 1337.4694,
"description": "fieldLength",
"details": []
}
]
}
]
}
]
},
{
"value": 2,
"description": "secondaryWeight",
"details": []
}
]
}
]
}
I think it's just a known issue with rescorer. If the window_size (which default to 10) of the rescorer is smaller than the requested size (from+size, in the example query 0+67) then all documents that appear after the window_size will skip the rescore and they will just apply the query_weight to the original score. Though we don't check if the document was part of the window when explain is called which is why the score within the explanation is not correct. It shows the score as if it were rescored even though the document might not have been rescored.
You should have the same score if you change the window_size to match the size+from of you query (67 ;) ).
I'll leave this issue open for now because there is a bug in the rescorer explain output but the fix is not trivial and I am not sure that it is worth fixing.
jimczi
on 19 Feb 2018
👍2
I think we should fix it: scores are explanations should always match.
jpountz
on 21 Feb 2018
Fair enough, I restored the adoptme tag. Also note that the score will not always match the explanations since we can have more than one rescorer so only the explanation of the last rescorer should match the final score of the hit.
jimczi
on 21 Feb 2018
To implement the correct explain in QueryRescorer explain function, we need to check if a document was within the window_size - N. This is difficult to check, as documents can be resorted, and the top N documents are not necessarily the one for which rescoring was applied (in case for example rescore_query_weight is negative, they may be at the bottom). We can potentially modify RescoreContext to save all docIds for which rescore was applied, and use this info during ExplainFetchSubPhase, but I am not sure if we want to go this round.
The simpler solution would be to remove window_size parameter from rescore API, and apply rescore for all docs between [from, from+size).
What do you think @jimczi @jpountz ?
mayya-sharipova
on 30 Apr 2018
The simpler solution would be to remove window_size parameter from rescore API, and apply rescore for all docs between [from, from+size).
The window_size is important because it ensures that pagination queries always rescore the same hits. If we remove it the pagination would rescore more documents on each page and this would yields different results (or duplicates) on each page.
We can potentially modify RescoreContext to save all docIds for which rescore was applied, and use this info during ExplainFetchSubPhase, but I am not sure if we want to go this round.
The window_size should be small (100-1000 or something) so I think that's ok and I don't see how we could do it differently.
jimczi
on 3 May 2018
👍1
Was this page helpful?
0 / 5 - 0 ratings
Related issues
abrahamduran
路
3Comments
makeyang
路
3Comments
dawi
路
3Comments
Praveen82
路
3Comments
jasontedor
路
3Comments
Most helpful comment
I think it's just a known issue with
rescorer. If thewindow_size(which default to10) of therescoreris smaller than the requested size (from+size, in the example query0+67) then all documents that appear after thewindow_sizewill skip the rescore and they will just apply thequery_weightto the original score. Though we don't check if the document was part of the window whenexplainis called which is why the score within the explanation is not correct. It shows the score as if it were rescored even though the document might not have been rescored.You should have the same score if you change the
window_sizeto match thesize+fromof you query (67;) ).I'll leave this issue open for now because there is a bug in the rescorer
explainoutput but the fix is not trivial and I am not sure that it is worth fixing.