I am new to Elastic Search and I wanted to ask how can I boost certain results from a query based on some conditional. Specifically, the items in Elastic Search are of the following format:
"name":"item1",
"desc":"desc1",
"brand":"brandname"
I want to make the score of certain items (those with "brand" = "nike") have lower score than the other brands. Is there a way to do that with the QueryBuilders classes that java has for elastic search?
Boosting query might be able to solve the problem. Alternatively, you can also use function_score queries for more flexibility.
Example with boosting query.
(1) Index few documents
curl -XPOST localhost:9200/idx/doc/1 -d '{"content": "foo"}' --header "Content-type:application/json"
curl -XPOST localhost:9200/idx/doc/2 -d '{"content": "bar"}' --header "Content-type:application/json"
curl -XPOST localhost:9200/idx/doc/3 -d '{"content": "foo fred bar"}' --header "Content-type:application/json"
(2) Query with a negative boost for bar to reduce score for bar
curl -XPOST localhost:9200/idx/_search?pretty -d '{
"query": {
"boosting": {
"positive": {
"match": {
"content": "foo bar fred"
}
},
"negative": {
"match": {
"content": "bar"
}
} , "negative_boost": 0.001
}
}
}' --header "Content-type:application/json"
{
"took" : 59,
"timed_out" : false,
"_shards" : {
"total" : 1,
"successful" : 1,
"skipped" : 0,
"failed" : 0
},
"hits" : {
"total" : {
"value" : 3,
"relation" : "eq"
},
"max_score" : 0.5619609,
"hits" : [
{
"_index" : "idx",
"_type" : "doc",
"_id" : "1",
"_score" : 0.5619609,
"_source" : {
"content" : "foo"
}
},
{
"_index" : "idx",
"_type" : "doc",
"_id" : "3",
"_score" : 0.0014472058,
"_source" : {
"content" : "foo fred bar"
}
},
{
"_index" : "idx",
"_type" : "doc",
"_id" : "2",
"_score" : 5.619609E-4,
"_source" : {
"content" : "bar"
}
}
]
}
}
To boost a certain value of a certain field (or vice versa), you can use a Function score query.
Check the official document Here.
Let me show you an example :
POST /goods/_search
{
"query": {
"function_score": {
"query": {
"match": {
"name": "shoes"
}
},
"functions": [
{
"filter": {
"term": {
"maker": "nike"
}
},
"weight": 0
}
],
"boost_mode": "multiply"
}
},
"sort": [
{
"_score": {
"order": "desc"
}
}
]
}
This query refers that "I'm looking for documents which match a keyword shoes in a name field and sort in reverse order of _score. If documents have a term nike in a maker field, however, I hope that the score is 0(cause multiplying any number by 0 equals 0)."
In Java code,
String indexName = "goods"
String nameField = "name";
String makerField = "maker";
String keyword = "shoes";
String unnecessaryMaker = "nike";
MatchQueryBuilder matchQueryBuilder = new MatchQueryBuilder(nameField, keyword);
FilterFunctionBuilder[] filterFunctions = {
new FunctionScoreQueryBuilder.FilterFunctionBuilder(
QueryBuilders.termQuery(makerField, unnecessaryMaker),
ScoreFunctionBuilders.weightFactorFunction(0)
)
};
FunctionScoreQueryBuilder functionScoreQueryBuilder = QueryBuilders.functionScoreQuery(matchQueryBuilder, filterFunctions);
functionScoreQueryBuilder.boostMode(CombineFunction.MULTIPLY);
SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
searchSourceBuilder.query(functionScoreQueryBuilder);
SearchRequest request = new SearchRequest();
request.indices(indexName);
request.source(searchSourceBuilder);
There are several boost_mode such as multiply(default), replace, sum, avg, max, min.
So you can apply this code to your needs.
Hope this is what you were looking for.
Related
I have a field called Description which is a text field and has data like:
This is a good thing for versions before 3.2 but bad for 3.5 and later
I want to run range query on this type of text. I know that for a field containing only Dates/Age(Numbers) or even String Ids, we can use queries like
{
"query": {
"range" : {
"age" : {
"gte" : 10,
"lte" : 20,
"boost" : 2.0
}
}
}
}
But i have a mixed field like mentioned above and I need to perform range query on that. Also, i cannot change the index structure. I can only perform queries or do some post processing after retrieving results. So anyone has any idea how to run this type of query, or even obtain my goal after getting results in the post processing? I am using Java.
I hope i fully understand what you are looking for.
I've managed to create a simple working example.
Mappings
Using char_group tokenizer:
The char_group tokenizer breaks text into terms whenever it encounters a character which is in a defined set. It is mostly useful for cases where a simple custom tokenization is desired, and the overhead of use of the pattern tokenizer is not acceptable.
Char Group Tokenizer
PUT my_index
{
"settings": {
"analysis": {
"analyzer": {
"my_analyzer": {
"type": "custom",
"tokenizer": "my_tokenizer"
}
},
"tokenizer": {
"my_tokenizer": {
"type": "char_group",
"tokenize_on_chars": [
"letter",
"whitespace"
]
}
}
}
},
"mappings": {
"properties": {
"text": {
"type": "text",
"fields": {
"digit": {
"type": "text",
"analyzer": "my_analyzer"
}
}
}
}
}
}
Post a few documents
PUT my_index/_doc/1
{
"text": "This is a good thing for versions before 3.2 but bad for 3.5 and later"
}
PUT my_index/_doc/2
{
"text": "This is a good thing for versions before 5 but bad for 6 and later"
}
Search Query
GET my_index/_search
{
"query": {
"range": {
"text.digit": {
"gte": 3.2,
"lte": 3.5
}
}
}
}
Results
"hits" : {
"total" : {
"value" : 1,
"relation" : "eq"
},
"max_score" : 1.0,
"hits" : [
{
"_index" : "my_index",
"_type" : "_doc",
"_id" : "1",
"_score" : 1.0,
"_source" : {
"text" : "This is a good thing for versions before 3.2 but bad for 3.5 and later"
}
}
]
}
Another Search Query
GET my_index/_search
{
"query": {
"range": {
"text.digit": {
"gt": 3.5
}
}
}
}
Results
"hits" : {
"total" : {
"value" : 1,
"relation" : "eq"
},
"max_score" : 1.0,
"hits" : [
{
"_index" : "my_index",
"_type" : "_doc",
"_id" : "2",
"_score" : 1.0,
"_source" : {
"text" : "This is a good thing for versions before 5 but bad for 6 and later"
}
}
]
}
Analyze Query
Play with the following query till you get the desired results.
It is already compatible to your example.
This is a good thing for versions before 3.2 but bad for 3.5 and later
POST _analyze
{
"tokenizer": {
"type": "char_group",
"tokenize_on_chars": [
"letter",
"whitespace"
]
},
"text": "This is a good thing for versions before 3.2 but bad for 3.5 and later"
}
Hope this helps
I have indexed sample documents in elasticsearch and trying to search using fuzzy query. But am not getting any results when am search by using Java fuzzy query api.
Please find my below mapping script :
PUT productcatalog
{
"settings": {
"analysis": {
"analyzer": {
"attr_analyzer": {
"type": "custom",
"tokenizer": "letter",
"char_filter": [
"html_strip"
],
"filter": ["lowercase", "asciifolding", "stemmer_minimal_english"]
}
},
"filter" : {
"stemmer_minimal_english" : {
"type" : "stemmer",
"name" : "minimal_english"
}
}
}
},
"mappings": {
"doc": {
"properties": {
"values": {
"type": "text",
"analyzer": "attr_analyzer"
},
"catalog_type": {
"type": "text"
},
"catalog_id":{
"type": "long"
}
}
}
}
}
Please find my sample data.
PUT productcatalog/doc/1
{
"catalog_id" : "343",
"catalog_type" : "series",
"values" : "Activa Rooftop, valves, VG3000, VG3000FS, butterfly, ball"
}
PUT productcatalog/doc/2
{
"catalog_id" : "12717",
"catalog_type" : "product",
"values" : "Activa Rooftop, valves"
}
Please find my search script :
GET productcatalog/_search
{
"query": {
"match" : {
"values" : {
"query" : " activa rooftop VG3000",
"operator" : "and",
"boost": 1.0,
"fuzziness": 2,
"prefix_length": 0,
"max_expansions": 100
}
}
}
}
Am getting the below results for the above query :
{
"took": 239,
"timed_out": false,
"_shards": {
"total": 5,
"successful": 5,
"skipped": 0,
"failed": 0
},
"hits": {
"total": 1,
"max_score": 0.970927,
"hits": [
{
"_index": "productcatalog",
"_type": "doc",
"_id": "1",
"_score": 0.970927,
"_source": {
"catalog_id": "343",
"catalog_type": "series",
"values": "Activa Rooftop, valves, VG3000, VG3000FS, butterfly, ball"
}
}
]
}
}
But if i use the below Java API for the same fuzzy search am not getting any results out of it.
Please find my below Java API query for fuzzy search :
QueryBuilder qb = QueryBuilders.boolQuery()
.must(QueryBuilders.fuzzyQuery("values", keyword).boost(1.0f).prefixLength(0).maxExpansions(100));
Update 1
I have tried with the below query
QueryBuilder qb = QueryBuilders.matchQuery(QueryBuilders.fuzzyQuery("values", keyword).boost(1.0f).prefixLength(0).maxExpansions(100));
But am not able to pass QueryBuilders inside matchQuery. Am getting this suggestion while am writing this query The method matchQuery(String, Object) in the type QueryBuilders is not applicable for the arguments (FuzzyQueryBuilder)
The mentioned java query is not a match query. It's a must query. you should use matchQuery instead of boolQuery().must(QueryBuilders.fuzzyQuery())
Update 1:
fuzzy query is a term query while match query is a full text query.
Also don't forget that in match query the default Operator is or operator which you should change it to and like your dsl query.
I am trying to retrieve the data from SearchResponse class with the above code:
SearchHits searchHits = searchResponse.getHits();
for (SearchHit searchHit : searchHits) {
SearchHitField title = searchHit.field("title");
System.out.println(title.getValue().toString());
}
But I get a null pointer exception in title.getValue() function. The "title" field is definitely there and I can verify that by printing the search response which gives the following output:
{
"took" : 13,
"timed_out" : false,
"_shards" : {
"total" : 5,
"successful" : 5,
"failed" : 0
},
"hits" : {
"total" : 1,
"max_score" : 1.0,
"hits" : [ {
"_index" : "myIndex",
"_type" : "myTye",
"_id" : "5c849b0f-d72d-4cc9-9b8c-e1201f888f94",
"_score" : 2.4181843,
"_source":{"esId":"100200153", "title":"Book 1"}
}
}
I know that I can retrieve the data with searchHit.getSource() but I am wondering why the above solution isn't working as well.
I think you have to specify .fields(fields) in the request to be able to access the fields part.
For example, if you have a query like this:
{
"query": {
"match_all": {}
}
}
you get in the hits section of the result some fields (_id, _type..., _source).
But, if you have something like this:
{
"query": {
"match_all": {}
},
"fields": ["my_field"]
}
you get back a different result:
"hits": {
"total": 2,
"max_score": 1,
"hits": [
{
"_index": "test_malformed",
"_type": "test",
"_id": "1",
"_score": 1,
"fields": {
"my_field": [
"whatever"
]
}
},
...
You notice there, in the hits you have fields where the field specified in the search request is being returned.
It looks like you are almost there. On each hit, instead of getting the title, get the _source object, then the title field from that source object.
I have a elasticsearch index with a field "aDate" (and lot of other fields) with the following mapping
"aDate" : {
"type" : "date",
"format" : "date_optional_time"
}
When i query for a document i get a result like
"aDate" : 1421179734000,
I know this is the epoch, the internal java/elasticsearch date format, but i want to have a result like:
"aDate" : "2015-01-13T20:08:54",
I play around with scripting
{
"query":{
"match_all":{
}
},
"script_fields":{
"aDate":{
"script":"if (!_source.aDate?.equals('null')) new java.text.SimpleDateFormat('yyyy-MM-dd\\'T\\'HH:mm:ss').format(new java.util.Date(_source.aDate));"
}
}
}
but it give strange results (script works basically, but aDate is the only field returned and _source is missing). This looks like
"hits": [{
"_index": "idx1",
"_type": "type2",
"_id": "8770",
"_score": 1.0,
"fields": {
"aDate": ["2015-01-12T17:15:47"]
}
},
I would prefer a solution without scripting if possible.
When you run a query in Elasticsearch you can request it to return the raw data, for example specifying fields:
curl -XGET http://localhost:9200/myindex/date-test/_search?pretty -d '
{
"fields" : "aDate",
"query":{
"match_all":{
}
}
}'
Will give you the date in the format that you originally stored it:
{
"_index" : "myindex",
"_type" : "date-test",
"_id" : "AUrlWNTAk1DYhbTcL2xO",
"_score" : 1.0,
"fields" : {
"aDate" : [ "2015-01-13T20:08:56" ]
}
}, {
"_index" : "myindex",
"_type" : "date-test",
"_id" : "AUrlQnFgk1DYhbTcL2xM",
"_score" : 1.0,
"fields" : {
"aDate" : [ 1421179734000 ]
}
It's not possible to change the date format unless you use a script.
curl -XGET http://localhost:9200/myindex/date-test/_search?pretty -d '
{
"query":{
"match_all":{ }
},
"script_fields":{
"aDate":{
"script":"use( groovy.time.TimeCategory ) { new Date( doc[\"aDate\"].value ) }"
}
}
}'
Will return:
{
"_index" : "myindex",
"_type" : "date-test",
"_id" : "AUrlWNTAk1DYhbTcL2xO",
"_score" : 1.0,
"fields" : {
"aDate" : [ "2015-01-13T20:08:56.000Z" ]
}
}, {
"_index" : "myindex",
"_type" : "date-test",
"_id" : "AUrlQnFgk1DYhbTcL2xM",
"_score" : 1.0,
"fields" : {
"aDate" : [ "2015-01-13T20:08:54.000Z" ]
}
}
To apply a format, append it as follows:
"script":"use( groovy.time.TimeCategory ){ new Date( doc[\"aDate\"].value ).format(\"yyyy-MM-dd\") }"
will return "aDate" : [ "2015-01-13" ]
To display the T, you'll need to use quotes but replace them with the Unicode equivalent:
"script":"use( groovy.time.TimeCategory ){ new Date( doc[\"aDate\"].value ).format(\"yyyy-MM-dd\u0027T\u0027HH:mm:ss\") }"
returns "aDate" : [ "2015-01-13T20:08:54" ]
To return script_fields and source
Use _source in your query to specify the fields you want to return:
curl -XGET http://localhost:9200/myindex/date-test/_search?pretty -d '
{ "_source" : "name",
"query":{
"match_all":{ }
},
"script_fields":{
"aDate":{
"script":"use( groovy.time.TimeCategory ) { new Date( doc[\"aDate\"].value ) }"
}
}
}'
Will return my name field:
"_source":{"name":"Terry"},
"fields" : {
"aDate" : [ "2015-01-13T20:08:56.000Z" ]
}
Using asterisk will return all fields, e.g.: "_source" : "*",
"_source":{"name":"Terry","aDate":1421179736000},
"fields" : {
"aDate" : [ "2015-01-13T20:08:56.000Z" ]
}
Since 5.0.0, es use Painless as script language: link
Try this (work in 6.3.2)
"script":"doc['aDate'].value.toString('yyyy-MM-dd HH:mm:ss')"
As LabOctoCat mentioned, Olly Cruickshank answer no longer works in elastic 2.2. I changed the script to:
"script":"new Date(doc['time'].value)"
You can format the date according to this.
Scripting it only computes the answer when the row is extracted. This is expensive, and keeps you from using any date-related search functions in Elasticsearch.
You should create an elasticsearch "date" field before inserting it. Looks like a java Date() object will do.
Thanks #Archon for your suggestion. I used your answer as a guide to remove the time element from a datetime field in Elasticsearch
{
"aggs": {
"grp_by_date": {
"terms": {
"size": 200,
"script": "doc['TransactionReconciliationsCreated'].value.toString('yyyy-MM-dd')"
}
}
}
}
If you use Elasticsearch 7, and want to display datetime in a specified timezone, you can request it like this
"query": {
"bool": {
"filter": [
{
"term": {
"client": {
"value": "iOS",
"boost": 1
}
}
}
],
"adjust_pure_negative": true,
"boost": 1
}
},
"script_fields": {
"time": {
"script": "ZonedDateTime input = doc['time'].value; input = input.withZoneSameInstant(ZoneId.of('Asia/Shanghai')); String output = input.format(DateTimeFormatter.ISO_ZONED_DATE_TIME); return output"
}
},
"_source": true,
return
{
...
"_source" : {
...
"time" : 1632903354213
...
},
"fields" : {
"time" : [
"2021-09-29T16:15:54.213+08:00[Asia/Shanghai]"
]
}
},
...
}
How can I have the tokens of a particular field returned in the result
For example, A GET request
curl -XGET 'http://localhost:9200/twitter/tweet/1'
returns
{
"_index" : "twitter",
"_type" : "tweet",
"_id" : "1",
"_source" : {
"user" : "kimchy",
"postDate" : "2009-11-15T14:12:12",
"message" : "trying out Elastic Search"
}
}
I would like to have the tokens of '_source.message' field included in the result
There is also another way to do it using the following script_fields script:
curl -H 'Content-Type: application/json' -XPOST 'http://localhost:9200/test-idx/_search?pretty=true' -d '{
"query" : {
"match_all" : { }
},
"script_fields": {
"terms" : {
"script": "doc[field].values",
"params": {
"field": "message"
}
}
}
}'
It's important to note that while this script returns the actual terms that were indexed, it also caches all field values and on large indices can use a lot of memory. So, on large indices, it might be more useful to retrieve field values from stored fields or source and reparse them again on the fly using the following MVEL script:
import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;
import java.io.StringReader;
// Cache analyzer for further use
cachedAnalyzer=(isdef cachedAnalyzer)?cachedAnalyzer:doc.mapperService().documentMapper(doc._type.value).mappers().indexAnalyzer();
terms=[];
// Get value from Fields Lookup
//val=_fields[field].values;
// Get value from Source Lookup
val=_source[field];
if(val != null) {
tokenStream=cachedAnalyzer.tokenStream(field, new StringReader(val));
CharTermAttribute termAttribute = tokenStream.addAttribute(CharTermAttribute);
while(tokenStream.incrementToken()) {
terms.add(termAttribute.toString())
};
tokenStream.close();
}
terms
This MVEL script can be stored as config/scripts/analyze.mvel and used with the following query:
curl 'http://localhost:9200/test-idx/_search?pretty=true' -d '{
"query" : {
"match_all" : { }
},
"script_fields": {
"terms" : {
"script": "analyze",
"params": {
"field": "message"
}
}
}
}'
If you mean the tokens that have been indexed you can make a terms facet on the message field. Increase the size value in order to get more entries back, or set to 0 to get all terms.
Lucene provides the ability to store the term vectors, but there's no way to have access to it with elasticsearch by now (as far as I know).
Why do you need that? If you only want to check what you're indexing you can have a look at the analyze api.
Nowadays, it's possible with the Term vectors API:
curl http://localhost:9200/twitter/_termvectors/1?fields=message
Result:
{
"_index": "twitter",
"_id": "1",
"_version": 1,
"found": true,
"took": 0,
"term_vectors": {
"message": {
"field_statistics": {
"sum_doc_freq": 4,
"doc_count": 1,
"sum_ttf": 4
},
"terms": {
"elastic": {
"term_freq": 1,
"tokens": [
{
"position": 2,
"start_offset": 11,
"end_offset": 18
}
]
},
"out": {
"term_freq": 1,
"tokens": [
{
"position": 1,
"start_offset": 7,
"end_offset": 10
}
]
},
"search": {
"term_freq": 1,
"tokens": [
{
"position": 3,
"start_offset": 19,
"end_offset": 25
}
]
},
"trying": {
"term_freq": 1,
"tokens": [
{
"position": 0,
"start_offset": 0,
"end_offset": 6
}
]
}
}
}
}
}
Note: Mapping types (here: tweets) have been removed in Elasticsearch 8.x (see migration guide).