# Synonym\_graph + match\_phrase: Unexpected high score due to the sum up of IDF of all matched synonym words **URL:** \ **Category:** Elasticsearch\ **Created:** [September 3, 2021, 9:28am UTC](https://discuss.elastic.co/t/synonym-graph-match-phrase-unexpected-high-score-due-to-the-sum-up-of-idf-of-all-matched-synonym-words/283260 "2021-09-03T09:28:35Z")\ **Posts on this page:** 2\ **Page:** 1
**Author:** ![Patrick\_N](https://avatars.discourse-cdn.com/v4/letter/p/8dc957/32.png) [@Patrick\_N](https://discuss.elastic.co/u/Patrick_N)\ **Post date:** [September 3, 2021, 9:28am UTC](https://discuss.elastic.co/t/synonym-graph-match-phrase-unexpected-high-score-due-to-the-sum-up-of-idf-of-all-matched-synonym-words/283260/1 "2021-09-03T09:28:35Z")
(ES 7.10) I use synonym\_graph `(usa, united states north america)` for the **search\_analyzer** of a field. When I perform a match\_phrase which contains `usa`, the score of any matching doc which contains the term `usa` will be **substantial** higher than the rest. ```auto GET test/_search { "explain": true, "query": { "bool": { "should": [ { "match_phrase": {"name": {"query": "usa"}} }, { "match_phrase": {"name": {"query": "happy boy"}} } ] } } } ``` On a closer look, it's because the IDF of that matched doc is the sum of the IDF of **all** the terms in the matched synonym (usa, united, states, north, america), instead of just the IDF of the matching term (usa). Explanation (I just extracted only the part for the 1st hit): ```auto { "_shard" : "[test][0]", "_index" : "test", "_type" : "_doc", "_id" : "1", "_score" : 6.9519825, "_source" : { "name" : "usa happy boy" }, "_explanation" : { "value" : 6.9519825, "description" : "sum of:", "details" : [ { "value" : 6.215455, "description" : "weight(spanOr([spanNear([name:united, name:states, name:north, name:america], 0, true), name:usa]) in 0) [PerFieldSimilarity], result of:", "details" : [ { "value" : 6.215455, "description" : "score(freq=1.0), computed as boost * idf * tf from:", "details" : [ { "value" : 2.2, "description" : "boost", }, { "value" : 6.019864, "description" : "idf, sum of:", "details" : [ { "value" : 1.2039728, "description" : "idf, computed as log(1 + (N - n + 0.5) / (n + 0.5)) from:", "details" : [ { "value" : 1, "description" : "n, number of documents containing term", }, { "value" : 4, "description" : "N, total number of documents with field", } ] }, { "value" : 1.2039728, "description" : "idf, computed as log(1 + (N - n + 0.5) / (n + 0.5)) from:", "details" : [ { "value" : 1, "description" : "n, number of documents containing term", }, { "value" : 4, "description" : "N, total number of documents with field", } ] }, { "value" : 1.2039728, "description" : "idf, computed as log(1 + (N - n + 0.5) / (n + 0.5)) from:", "details" : [ { "value" : 1, "description" : "n, number of documents containing term", }, { "value" : 4, "description" : "N, total number of documents with field", } ] }, { "value" : 1.2039728, "description" : "idf, computed as log(1 + (N - n + 0.5) / (n + 0.5)) from:", "details" : [ { "value" : 1, "description" : "n, number of documents containing term", }, { "value" : 4, "description" : "N, total number of documents with field", } ] }, { "value" : 1.2039728, "description" : "idf, computed as log(1 + (N - n + 0.5) / (n + 0.5)) from:", "details" : [ { "value" : 1, "description" : "n, number of documents containing term", }, { "value" : 4, "description" : "N, total number of documents with field", } ] } ] }, { "value" : 0.46931404, "description" : "tf, computed as freq / (freq + k1 * (1 - b + b * dl / avgdl)) from:", ("details of tf removed) } ] } ] }, { "value" : 0.73652726, "description" : """weight(name:"happy boy" in 0) [PerFieldSimilarity], result of:""", "details" : [ { "value" : 0.73652726, "description" : "score(freq=1.0), computed as boost * idf * tf from:", "details" : [ { "value" : 2.2, "description" : "boost", }, { "value" : 0.7133499, "description" : "idf, sum of:", "details" : [ { "value" : 0.35667494, "description" : "idf, computed as log(1 + (N - n + 0.5) / (n + 0.5)) from:", "details" : [ { "value" : 3, "description" : "n, number of documents containing term" }, { "value" : 4, "description" : "N, total number of documents with field" } ] }, { "value" : 0.35667494, "description" : "idf, computed as log(1 + (N - n + 0.5) / (n + 0.5)) from:", "details" : [ { "value" : 3, "description" : "n, number of documents containing term" }, { "value" : 4, "description" : "N, total number of documents with field" } ] } ] }, { "value" : 0.46931404, "description" : "tf, computed as freq / (freq + k1 * (1 - b + b * dl / avgdl)) from:", (details of tf removed) } ] } ] } ] } }, ``` This is causing a problem, because in my program, I am search multiple fields, and this unexpected high IDF for **any doc which matches a synonym** is causing some ranking problems. E.g. if I do the below search, I won't run into that high IDF problem for the 1st matched doc ```auto GET test/_search { "explain": true, "query": { "bool": { "should": [ { "match_phrase": {"name": {"query": "germany"}} }, { "match_phrase": {"name": {"query": "happy boy"}} } ] } } } ``` Since only the term `usa` from the search matches, can we count only the IDF of that term? Is that possible? Setup: ```auto DELETE test PUT test { "settings": { "number_of_replicas": 0, "number_of_shards": 1, "analysis": { "filter": { "synonyms": { "type": "synonym_graph", "synonyms": [ "usa, united states north america" ] } }, "analyzer": { "synonym_analyzer": { "tokenizer": "whitespace", "filter": [ "synonyms" ] } } } }, "mappings": { "properties": { "name": { "type": "text", "analyzer": "whitespace", "search_analyzer": "synonym_analyzer" } } } } #### Add testing doc PUT test/_doc/1 { "name" : "usa happy boy" } PUT test/_doc/2 { "name" : "orange happy boy" } PUT test/_doc/3 { "name" : "germany happy boy" } PUT test/_doc/4 { "name" : "united states north america" } ``` ---
**Author:** ![system](https://us1.discourse-cdn.com/elastic/original/3X/1/a/1ac57faf039f6b580b3f104ef42a2a89e41014de.png) [@system](https://discuss.elastic.co/u/system)\ **Post date:** [October 1, 2021, 9:29am UTC](https://discuss.elastic.co/t/synonym-graph-match-phrase-unexpected-high-score-due-to-the-sum-up-of-idf-of-all-matched-synonym-words/283260/2 "2021-10-01T09:29:01Z")
This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.