Full-Text Search Internals Philipp Krenn @xeraa

Developer

Who uses a Database?

Who uses Search?

Store

Apache Lucene Elasticsearch

Example These are <em>not</em> the droids you are looking for.

html_strip Char Filter These are not the droids you are looking for.

standard Tokenizer These are not the droids you looking for are

lowercase Token Filter these are not the droids looking for you are

stop Token Filter droids you looking

snowball Token Filter droid you look

Setup

https://cloud.elastic.co

Docker Compose --version: '2' services: kibana: image: docker.elastic.co/kibana/kibana:$ELASTIC_VERSION links: - elasticsearch ports: - 5601:5601 elasticsearch: image: docker.elastic.co/elasticsearch/elasticsearch:$ELASTIC_VERSION volumes: - esdata1:/usr/share/elasticsearch/data ports: - 9200:9200 volumes: esdata1: driver: local

Analyze

GET /_analyze { "analyzer": "english", "text": "These are not the droids you are looking for." }

{ } "tokens": [ { "token": "droid", "start_offset": 18, "end_offset": 24, "type": "<ALPHANUM>", "position": 4 }, { "token": "you", "start_offset": 25, "end_offset": 28, "type": "<ALPHANUM>", "position": 5 }, ... ]

GET /_analyze { "char_filter": [ "html_strip" ], "tokenizer": "standard", "filter": [ "lowercase", "stop", "snowball" ], "text": "These are <em>not</em> the droids you are looking for." }

{ } "tokens": [ { "token": "droid", "start_offset": 27, "end_offset": 33, "type": "<ALPHANUM>", "position": 4 }, { "token": "you", "start_offset": 34, "end_offset": 37, "type": "<ALPHANUM>", "position": 5 }, ... ]

Stop Words a an and are as at be but by for if in into is it no not of on or such that the their then there these they this to was will with https://github.com/apache/lucene-solr/blob/master/lucene/ analysis/common/src/java/org/apache/lucene/analysis/en/ EnglishAnalyzer.java#L44-L50

Always Use Stop Words?

To be, or not to be.

Languages Arabic, Armenian, Basque, Brazilian, Bulgarian, Catalan, CJK, Czech, Danish, Dutch, English, Finnish, French, Galician, German, Greek, Hindi, Hungarian, Indonesian, Irish, Italian, Latvian, Lithuanian, Norwegian, Persian, Portuguese, Romanian, Russian, Sorani, Spanish, Swedish, Turkish, Thai

Language Rules English: Philipp's → philipp French: l'église → eglis German: äußerst → ausserst

More Language Plugins Core: ICU (Asian languages), Kuromoji (advanced Japanese), Phonetic, SmartCN, Stempel (Polish), Ukrainian Community: Hebrew, Vietnamese, Network Address Analysis, String2Integer,...

German GET /_analyze { "analyzer": "german", "text": "Das sind nicht die Droiden, nach denen du suchst." }

{ } "tokens": [ { "token": "droid", "start_offset": 19, "end_offset": 26, "type": "<ALPHANUM>", "position": 4 }, { "token": "den", "start_offset": 33, "end_offset": 38, "type": "<ALPHANUM>", "position": 6 }, { "token": "such", "start_offset": 42, "end_offset": 48, "type": "<ALPHANUM>", "position": 8 } ]

German with the English Analyzer da sind nicht die droiden denen du suchst nach

German Stop Words https://github.com/apache/lucene-solr/blob/master/lucene/ analysis/common/src/resources/org/apache/lucene/analysis/ snowball/german_stop.txt

Detect Languages https://github.com/spinscale/ elasticsearch-ingest-langdetect

PUT _ingest/pipeline/langdetect-pipeline { "description": "A pipeline to detect languages", "processors": [ { "langdetect" : { "field" : "quote", "target_field" : "language" } } ] }

POST _ingest/pipeline/langdetect-pipeline/_simulate { "docs": [ { "_source": { "quote": "Das sind nicht die Droiden, nach denen du suchst." } } ] }

{ } "docs": [ { "doc": { "_index": "_index", "_type": "_type", "_id": "_id", "_source": { "language": "de", "quote": "Das sind nicht die Droiden, nach denen du suchst." }, "_ingest": { "timestamp": "2018-10-26T00:06:42.320613Z" } } } ]

Phonetic GET /_analyze { "tokenizer": "standard", "filter": [ { "type": "phonetic", "encoder": "beider_morse", "languageset": "any" } ], "text": "These are not the droids you are looking for." }

Phonetic ... drDts drits drots loknk... iou ari ori

Another Example Obi-Wan never told you what happened to your father.

Another Example obi wan never told you what happen your father

Another Example <b>No</b>. I am your father.

Another Example i am your father

Inverted Index am droid father happen i look never obi told wan what you your ID 1 0 1[4] 0 0 0 1[7] 0 0 0 0 0 1[5] 0 ID 2 0 0 1[9] 1[6] 0 0 1[2] 1[0] 1[3] 1[1] 1[5] 1[4] 1[8] ID 3 1[2] 0 1[4] 0 1[1] 0 0 0 0 0 0 0 1[3]

To / The Index

PUT /starwars { "settings": { "number_of_shards": 1, "analysis": { "filter": { "my_synonym_filter": { "type": "synonym", "synonyms": [ "father,dad", "droid => droid,machine" ] } },

}, } "analyzer": { "my_analyzer": { "char_filter": [ "html_strip" ], "tokenizer": "standard", "filter": [ "lowercase", "stop", "snowball", "my_synonym_filter" ] } }

} "mappings": { "_doc": { "properties": { "quote": { "type": "text", "analyzer": "my_analyzer" } } } }

Synonyms Index synonym or query time synonym_graph

GET /starwars/_mapping GET /starwars/_settings

PUT /starwars/_doc/1 { "quote": "These are <em>not</em> the droids you are looking for." } PUT /starwars/_doc/2 { "quote": "Obi-Wan never told you what happened to your father." } PUT /starwars/_doc/3 { "quote": "<b>No</b>. I am your father." }

GET /starwars/_doc/1 GET /starwars/_doc/1/_source

Multi Lingual Index PUT /starwars_en/_doc/1 Type Field { "quote_en": "...", "quote_de": "..." }

PS: Single Type per Index

Search

POST /starwars/_search { "query": { "match_all": { } } }

GET vs POST

{ "took": 1, "timed_out": false, "_shards": { "total": 5, "successful": 5, "failed": 0 }, "hits": { "total": 3, "max_score": 1, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "2", "_score": 1, "_source": { "quote": "Obi-Wan never told you what happened to your father." } }, ...

POST /starwars/_search { "query": { "match": { "quote": "droid" } } }

{ } "took": 2, "timed_out": false, "_shards": { "total": 5, "successful": 5, "failed": 0 }, "hits": { "total": 1, "max_score": 0.39556286, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "1", "_score": 0.39556286, "_source": { "quote": "These are <em>not</em> the droids you are looking for." } } ] }

POST /starwars/_search { "query": { "match": { "quote": "dad" } } }

... "hits": { "total": 2, "max_score": 0.41913947, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "3", "_score": 0.41913947, "_source": { "quote": "<b>No</b>. I am your father." } }, { "_index": "starwars", "_type": "_doc", "_id": "2", "_score": 0.39291072, "_source": { "quote": "Obi-Wan never told you what happened to your father." } } ] } }

POST /starwars/_doc/0/_explain { "query": { "match": { "quote": "dad" } } }

{ } "_index": "starwars", "_type": "_doc", "_id": "0", "matched": false

POST /starwars/_doc/1/_explain { "query": { "match": { "quote": "dad" } } }

{ } "_index": "starwars", "_type": "_doc", "_id": "1", "matched": false, "explanation": { "value": 0, "description": "no matching term", "details": [] }

POST /starwars/_doc/2/_explain { "query": { "match": { "quote": "dad" } } }

{ "_index": "starwars", "_type": "_doc", "_id": "2", "matched": true, "explanation": { ...

POST /starwars/_search { "query": { "match": { "quote": "machine" } } }

{ } "took": 2, "timed_out": false, "_shards": { "total": 1, "successful": 1, "skipped": 0, "failed": 0 }, "hits": { "total": 1, "max_score": 1.2499592, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "1", "_score": 1.2499592, "_source": { "quote": "These are <em>not</em> the droids you are looking for." } } ] }

POST /starwars/_search { "query": { "match_phrase": { "quote": "I am your father" } } }

{ } "took": 3, "timed_out": false, "_shards": { "total": 5, "successful": 5, "failed": 0 }, "hits": { "total": 1, "max_score": 1.5665855, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "3", "_score": 1.5665855, "_source": { "quote": "<b>No</b>. I am your father." } } ] }

POST /starwars/_search { "query": { "match_phrase": { "quote": { "query": "I am father", "slop": 1 } } } }

{ } "took": 16, "timed_out": false, "_shards": { "total": 5, "successful": 5, "failed": 0 }, "hits": { "total": 1, "max_score": 0.8327639, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "3", "_score": 0.8327639, "_source": { "quote": "<b>No</b>. I am your father." } } ] }

POST /starwars/_search { "query": { "match_phrase": { "quote": { "query": "I am not your father", "slop": 1 } } } }

{ } "took": 5, "timed_out": false, "_shards": { "total": 5, "successful": 5, "failed": 0 }, "hits": { "total": 1, "max_score": 1.0409548, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "3", "_score": 1.0409548, "_source": { "quote": "<b>No</b>. I am your father." } } ] }

POST /starwars/_search { "query": { "match": { "quote": { "query": "van", "fuzziness": "AUTO" } } } }

{ } "took": 14, "timed_out": false, "_shards": { "total": 5, "successful": 5, "failed": 0 }, "hits": { "total": 1, "max_score": 0.18155496, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "2", "_score": 0.18155496, "_source": { "quote": "Obi-Wan never told you what happened to your father." } } ] }

POST /starwars/_search { "query": { "match": { "quote": { "query": "ovi-van", "fuzziness": 1 } } } }

{ } "took": 109, "timed_out": false, "_shards": { "total": 5, "successful": 5, "failed": 0 }, "hits": { "total": 1, "max_score": 0.3798467, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "2", "_score": 0.3798467, "_source": { "quote": "Obi-Wan never told you what happened to your father." } } ] }

FuzzyQuery History http://blog.mikemccandless.com/2011/03/lucenes-fuzzyquery-is-100-times-faster.html Before: Brute force Now: Levenshtein Automaton

http://blog.notdot.net/2010/07/Damn-Cool-Algorithms-Levenshtein-Automata

SELECT * FROM starwars WHERE quote LIKE "?an" OR quote LIKE "V?n" OR quote LIKE "Va?"

Score

Term Frequency / Inverse Document Frequency (TF/IDF) Search one term

BM25 Default in Elasticsearch 5.0 https://speakerdeck.com/elastic/improved-text-scoring-withbm25

Term Frequency

Inverse Document Frequency

Field-Length Norm

POST /starwars/_search?explain=true { "query": { "match": { "quote": "father" } } }

... "_explanation": { "value": 0.41913947, "description": "weight(Synonym(quote:dad quote:father) in 0) [PerFieldSimilarity], result of:", "details": [ { "value": 0.41913947, "description": "score(doc=0,freq=2.0 = termFreq=2.0\n), product of:", "details": [ { "value": 0.2876821, "description": "idf(docFreq=1, docCount=1)", "details": [] }, { "value": 1.4569536, "description": "tfNorm, computed from:", "details": [ { "value": 2, "description": "termFreq=2.0", "details": [] }, ...

Score 0.41913947: i am your father 0.39291072: obi wan never told what happen your father you

Vector Space Model Search multiple terms

Search your father

Coordination Factor Reward multiple terms

Search for 3 terms 1 term: 2 terms: 3 terms:

Practical Scoring Function Putting it all together

score(q,d) = queryNorm(q) · coord(q,d) · ∑ ( tf(t in d) · idf(t)² · t.getBoost() · norm(t,d) ) (t in q)

Function Score Script, weight, random, field value, decay (geo or date)

POST /starwars/_search { "query": { "function_score": { "query": { "match": { "quote": "father" } }, "random_score": {} } } }

Compare Scores "100% perfect" vs a "50%" match

Don't do this. Seriously. Stop trying to think about your problem this way, it's not going to end well. — https://wiki.apache.org/lucene-java/ ScoresAsPercentages

GET /starwars/_analyze { "analyzer" : "my_analyzer", "text": "These are my father's machines." }

{ "tokens": [ { "token": "my", "start_offset": 10, "end_offset": 12, "type": "<ALPHANUM>", "position": 2 }, { "token": "father", "start_offset": 13, "end_offset": 21, "type": "<ALPHANUM>", "position": 3 }, { "token": "dad", "start_offset": 13, "end_offset": 21, "type": "SYNONYM", "position": 3 }, { "token": "machin", "start_offset": 22, "end_offset": 30, "type": "<ALPHANUM>", "position": 4 } ] }

PUT /starwars/_doc/4 { "quote": "These are my father's machines." }

POST /starwars/_search { "query": { "match": { "quote": "my father machine" } } }

"hits": { "total": 4, "max_score": 2.92523, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "4", "_score": 2.92523, "_source": { "quote": "These are my father's machines." } }, { "_index": "starwars", "_type": "_doc", "_id": "1", "_score": 0.8617505, "_source": { "quote": "These are <em>not</em> the droids you are looking for." } }, ...

2.92523 == 100%

DELETE /starwars/_doc/4 POST /starwars/_search { "query": { "match": { "quote": "my father machine" } } }

"hits": { "total": 3, "max_score": 1.2499592, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "1", "_score": 1.2499592, "_source": { "quote": "These are <em>not</em> the droids you are looking for." } }, ...

1.2499592 == 43% or 100%?

PUT /starwars/_doc/4 { "quote": "These droids are my father's father's machines." } POST /starwars/_search { "query": { "match": { "quote": "my father machine" } } }

"hits": { "total": 4, "max_score": 3.0068164, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "4", "_score": 3.0068164, "_source": { "quote": "These droids are my father's father's machines." } }, { "_index": "starwars", "_type": "_doc", "_id": "1", "_score": 0.89701396, "_source": { "quote": "These are <em>not</em> the droids you are looking for." } }, ...

3.0068164 == 103%?

PS: Shards Default? Effect on IDF?

Distributed Frequency Search GET starwars/_search?search_type=dfs_query_then_fetch { ... }

Don’t use dfs_query_then_fetch in production. It really isn’t required. — https://www.elastic.co/guide/en/elasticsearch/ guide/current/relevance-is-broken.html

More Search

Highlighting

POST /starwars/_search { "query": { "match": { "quote": "father" } }, "highlight": { "type": "unified", "pre_tags": [ "<tag>" ], "post_tags": [ "</tag>" ], "fields": { "quote": {} } } }

... "hits": { "total": 3, "max_score": 0.631961, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "4", "_score": 0.631961, "_source": { "quote": "These droids are my father's father's machines." }, "highlight": { "quote": [ "These droids are my <tag>father's</tag> <tag>father's</tag> machines." ] } }, ...

Boolean Queries must must_not should filter

POST /starwars/_search { "query": { "bool": { "must": { "match": { "quote": "father" } }, "should": [ { "match": { "quote": "your" } }, { "match": { "quote": "obi" } } ] } } }

... "hits": { "total": 3, "max_score": 2.117857, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "2", "_score": 2.117857, "_source": { "quote": "Obi-Wan never told you what happened to your father." } }, { "_index": "starwars", "_type": "_doc", "_id": "3", "_score": 1.3856719, "_source": { "quote": "<b>No</b>. I am your father." } }, ...

POST /starwars/_search { "query": { "bool": { "filter": { "match": { "quote": "father" } }, "should": [ { "match": { "quote": "your" } }, { "match": { "quote": "obi" } } ] } } }

... "hits": { "total": 3, "max_score": 1.6694657, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "2", "_score": 1.6694657, "_source": { "quote": "Obi-Wan never told you what happened to your father." } }, { "_index": "starwars", "_type": "_doc", "_id": "3", "_score": 0.8317767, "_source": { "quote": "<b>No</b>. I am your father." } },

Named Queries & minimum_should_match

POST /starwars/_search { "query": { "bool": { "must": { "match": { "quote": "father" } }, "should": [ { "match": { "quote": { "query": "your", "_name": "quote-your" } } }, { "match": { "quote": { "query": "obi", "_name": "quote-obi" } } }, { "match": { "quote": { "query": "droid", "_name": "quote-droid" } } } ], "minimum_should_match": 2 } } }

... "hits": { "total": 1, "max_score": 2.117857, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "2", "_score": 2.117857, "_source": { "quote": "Obi-Wan never told you what happened to your father." }, "matched_queries": [ "quote-obi", "quote-your" ] } ] } }

Boosting >1 increase, <1 decrease, <0 punish <0 removed in 7.0

POST /starwars/_search { "query": { "bool": { "must": { "match": { "quote": "father" } }, "should": [ { "match": { "quote": "your" } }, { "match": { "quote": { "query": "obi", "boost": 3 } } } ] } } }

... "hits": { "total": 3, "max_score": 4.2368493, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "2", "_score": 4.2368493, "_source": { "quote": "Obi-Wan never told you what happened to your father." } }, { "_index": "starwars", "_type": "_doc", "_id": "3", "_score": 1.3856719, "_source": { "quote": "<b>No</b>. I am your father." } }, ...

Search for father but prefer father father

POST /starwars/_search { "query": { "bool": { "must": { "match": { "quote": "father father" } } } } }

... "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "4", "_score": 1.263922, "_source": { "quote": "These droids are my father's father's machines." } }, { "_index": "starwars", "_type": "_doc", "_id": "3", "_score": 1.1077905, "_source": { "quote": "<b>No</b>. I am your father." } },

POST /starwars/_search { "query": { "bool": { "must": { "match": { "quote": "father" } }, "should": { "match_phrase": { "quote": "father father" } } } } }

... "hits": { "total": 3, "max_score": 9.146545, "hits": [ { "_index": "starwars", "_type": "_doc", "_id": "4", "_score": 9.146545, "_source": { "quote": "These droids are my father's father's machines." } }, { "_index": "starwars", "_type": "_doc", "_id": "3", "_score": 1.0454913, "_source": { "quote": "<b>No</b>. I am your father." } }, ...

Suggestion Suggest a similar text _search end point _suggest deprecated since 5.0

POST /starwars/_search { "query": { "match": { "quote": "drui" } }, "suggest": { "my_suggestion" : { "text" : "drui", "term" : { "field" : "quote" } } } }

... "hits": { "total": 0, "max_score": null, "hits": [] }, "suggest": { "my_suggestion": [ { "text": "drui", "offset": 0, "length": 4, "options": [ { "text": "droid", "score": 0.5, "freq": 1 } ] } ] } }

Multiple Suggesters term phrase completion context

NGram Partial matches Edge Gram

GET /_analyze { "char_filter": [ "html_strip" ], "tokenizer": { "type": "ngram", "min_gram": "3", "max_gram": "3", "token_chars": [ "letter" ] }, "filter": [ "lowercase" ], "text": "These are <em>not</em> the droids you are looking for." }

{ "tokens": [ { "token": "the", "start_offset": 0, "end_offset": 3, "type": "word", "position": 0 }, { "token": "hes", "start_offset": 1, "end_offset": 4, "type": "word", "position": 1 }, { "token": "ese", "start_offset": 2, "end_offset": 5, "type": "word", "position": 2 }, { "token": "are", "start_offset": 6, "end_offset": 9, "type": "word", "position": 3 }, ...

GET /_analyze { "char_filter": [ "html_strip" ], "tokenizer": { "type": "edge_ngram", "min_gram": "1", "max_gram": "3", "token_chars": [ "letter" ] }, "filter": [ "lowercase" ], "text": "These are <em>not</em> the droids you are looking for." }

{ "tokens": [ { "token": "t", "start_offset": 0, "end_offset": 1, "type": "word", "position": 0 }, { "token": "th", "start_offset": 0, "end_offset": 2, "type": "word", "position": 1 }, { "token": "the", "start_offset": 0, "end_offset": 3, "type": "word", "position": 2 }, { "token": "a", "start_offset": 6, "end_offset": 7, "type": "word", "position": 3 }, { "token": "ar", "start_offset": 6, "end_offset": 8, "type": "word", "position": 4 }, ...

Combining Analyzers Reindex Store multiple times Tune BM25 Combine scores

BM25 Revisited

https://www.elastic.co/blog/practical-bm25-part-2-the-bm25algorithm-and-its-variables

b field length amplification k1 term frequency saturation Default 0.75 Default 1.2

PUT /starwars_v42 { "settings": { "number_of_shards": 1, "index": { "similarity": { "default": { "type": "BM25", "b": 0, "k1": 0 } } },

"analysis": { "filter": { "my_synonym_filter": { "type": "synonym", "synonyms": [ "father,dad", "droid => droid,machine" ] }, "my_ngram_filter": { "type": "ngram", "min_gram": "3", "max_gram": "3", "token_chars": [ "letter" ] } },

"analyzer": { "my_lowercase_analyzer": { "char_filter": [ "html_strip" ], "tokenizer": "whitespace", "filter": [ "lowercase" ] }, "my_full_analyzer": { "char_filter": [ "html_strip" ], "tokenizer": "standard", "filter": [ "lowercase", "stop", "snowball", "my_synonym_filter" ] },

}, } } "my_ngram_analyzer": { "char_filter": [ "html_strip" ], "tokenizer": "whitespace", "filter": [ "lowercase", "stop", "my_ngram_filter" ] }

"mappings": { "_doc": { "properties": { "quote": { "type": "text", "fields": { "lowercase": { "type": "text", "analyzer": "my_lowercase_analyzer" }, "full": { "type": "text", "analyzer": "my_full_analyzer" }, "ngram": { "type": "text", "analyzer": "my_ngram_analyzer" } } } } } } }

POST /_reindex { "source": { "index": "starwars" }, "dest": { "index": "starwars_v42" } }

Aliases Atomic remove and add Point to multiple indices (read-only)

PUT _alias { "actions": [ { "add": { "index": "starwars_v42", "alias": "starwars_extended" } } ] }

POST /starwars/_search { "query": { "match": { "quote": "droid" } } }

"hits": [ { "_index": "starwars", "_type": "_doc", "_id": "4", "_score": 1.1533037, "_source": { "quote": "These droids are my father's father's machines." } }, { "_index": "starwars", "_type": "_doc", "_id": "1", "_score": 1.1295731, "_source": { "quote": "These are <em>not</em> the droids you are looking for." } } ]

POST /starwars_extended/_search { "query": { "match": { "quote.full": "droid" } } }

"hits": [ { "_index": "starwars_v42", "_type": "_doc", "_id": "1", "_score": 0.6931472, "_source": { "quote": "These are <em>not</em> the droids you are looking for." } }, { "_index": "starwars_v42", "_type": "_doc", "_id": "4", "_score": 0.6931472, "_source": { "quote": "These droids are my father's father's machines." } } ]

There are no "best" b and k1 values

POST /starwars_extended/_search?explain=true { "query": { "multi_match": { "query": "obiwan", "fields": [ "quote", "quote.lowercase", "quote.full", "quote.ngram" ], "type": "most_fields" } } }

... "hits": { "total": 1, "max_score": 0.4912064, "hits": [ { "_shard": "[starwars_v42][2]", "_node": "BCDwzJ4WSw2dyoGLTzwlqw", "_index": "starwars_v42", "_type": "_doc", "_id": "2", "_score": 0.4912064, "_source": { "quote": "Obi-Wan never told you what happened to your father." }, ...

Whitespace Tokenizer "weight( Synonym(quote.ngram:biw quote.ngram:iwa quote.ngram:obi quote.ngram:wan) in 0) [PerFieldSimilarity], result of:"

POST /starwars_extended/_search { "query": { "multi_match": { "query": "you", "fields": [ "quote", "quote.lowercase^5", "quote.full", "quote.ngram" ], "type": "best_fields" } } }

"hits": [ { "_index": "starwars_v42", "_type": "_doc", "_id": "1", "_score": 3.465736, "_source": { "quote": "These are <em>not</em> the droids you are looking for." } }, { "_index": "starwars_v42", "_type": "_doc", "_id": "2", "_score": 3.465736, "_source": { "quote": "Obi-Wan never told you what happened to your father." } }, { "_index": "starwars_v42", "_type": "_doc", "_id": "3", "_score": 0.35667494, "_source": { "quote": "<b>No</b>. I am your father." } } ]

Multi Match Type best_fields Score of the best field (default) cross_fields All terms in at least one field most_fields Score sum of all fields phrase

Different Analyzers for Indexing and Searching Per query In the mapping

POST /starwars_extended/_search { "query": { "match": { "quote.ngram": { "query": "the", "analyzer": "standard" } } } }

... "hits": [ { "_index": "starwars_extended", "_type": "_doc", "_id": "2", "_score": 0.38254172, "_source": { "quote": "Obi-Wan never told you what happened to your father." } }, { "_index": "starwars_extended", "_type": "_doc", "_id": "3", "_score": 0.36165747, "_source": { "quote": "<b>No</b>. I am your father." } } ] ...

Edge Gram vs Trigram Test a setting before adding a field

Shingle Token Filter Shingles (token ngrams) from a token stream

POST /starwars_extended/_close PUT /starwars_extended/_settings { "index": { "similarity": { "default": { "type": "BM25", "b": null, "k1": null } } },

"analysis": { "filter": { "my_edgegram_filter": { "type": "edge_ngram", "min_gram": 3, "max_gram": 10 }, "my_shingle_filter": { "type": "shingle", "min_shingle_size": 2, "max_shingle_size": 2 } },

"analyzer": { "my_edgegram_analyzer": { "char_filter": [ "html_strip" ], "tokenizer": "standard", "filter": [ "lowercase", "my_edgegram_filter" ] },

} } } "my_shingle_analyzer": { "type": "custom", "tokenizer": "standard", "filter": [ "lowercase", "my_shingle_filter" ] } POST /starwars_extended/_open

GET starwars_extended/_analyze { "text": "Father", "analyzer": "my_edgegram_analyzer" }

{ } "tokens": [ { "token": "fat", "start_offset": 0, "end_offset": 6, "type": "<ALPHANUM>", "position": 0 }, { "token": "fath", "start_offset": 0, "end_offset": 6, "type": "<ALPHANUM>", "position": 0 }, { "token": "fathe", "start_offset": 0, "end_offset": 6, "type": "<ALPHANUM>", "position": 0 }, { "token": "father", "start_offset": 0, "end_offset": 6, "type": "<ALPHANUM>", "position": 0 } ]

PUT /starwars_extended/_doc/_mapping { "properties": { "quote": { "type": "text", "fields": { "edgegram": { "type": "text", "analyzer": "my_edgegram_analyzer", "search_analyzer": "standard" }, "shingle": { "type": "text", "analyzer": "my_shingle_analyzer" } } } } }

PUT /starwars_extended/_doc/5 { "quote": "I find your lack of faith disturbing." } PUT /starwars_extended/_doc/6 { "quote": "That... is your failure." }

GET /starwars_extended/_doc/5/_termvectors { "fields": [ "quote.edgegram" ], "offsets": true, "payloads": true, "positions": true, "term_statistics": true, "field_statistics": true }

{ "_index": "starwars_v42", "_type": "_doc", "_id": "5", "_version": 1, "found": true, "took": 3, "term_vectors": { "quote.edgegram": { "field_statistics": { "sum_doc_freq": 26, "doc_count": 2, "sum_ttf": 26 }, "terms": { "dis": { "doc_freq": 1, "ttf": 1, "term_freq": 1, "tokens": [ { "position": 6, "start_offset": 26, "end_offset": 36 } ] }, "dist": { "doc_freq": 1, "ttf": 1, ...

POST /starwars_extended/_search { "query": { "match": { "quote": "fail" } } }

POST /starwars_extended/_search { "query": { "match": { "quote.lowercase": "fail" } } }

POST /starwars_extended/_search { "query": { "match": { "quote.full": "fail" } } }

POST /starwars_extended/_search { "query": { "match": { "quote.ngram": "fail" } } }

"hits": [ { "_index": "starwars_v42", "_type": "_doc", "_id": "6", "_score": 1.8400999, "_source": { "quote": "That... is your failure." } }, { "_index": "starwars_v42", "_type": "_doc", "_id": "5", "_score": 1.442779, "_source": { "quote": "I find your lack of faith disturbing." } } ]

POST /starwars_extended/_search { "query": { "match": { "quote.edgegram": "fail" } } }

"hits": [ { "_index": "starwars_v42", "_type": "_doc", "_id": "6", "_score": 1.0114291, "_source": { "quote": "That... is your failure." } } ]

Updating Missing Fields Expensive

POST /starwars_extended/_update_by_query { "query": { "bool": { "must_not": { "exists": { "field": "quote.edgegram" } } } } }

Shingles: Context Should Matter

POST /starwars_extended/_search { "query": { "bool": { "must": { "match": { "quote.lowercase": "these droids are" } } } } }

"hits": [ { "_index": "starwars_v42", "_type": "_doc", "_id": "1", "_score": 2.1837702, "_source": { "quote": "These are <em>not</em> the droids you are looking for." } }, { "_index": "starwars_v42", "_type": "_doc", "_id": "4", "_score": 2.137744, "_source": { "quote": "These droids are my father's father's machines." } } ]

POST /starwars_extended/_search { "query": { "bool": { "must": { "match": { "quote.shingle": "these droids are" } } } } }

"hits": [ { "_index": "starwars_v42", "_type": "_doc", "_id": "4", "_score": 3.1811738, "_source": { "quote": "These droids are my father's father's machines." } }, { "_index": "starwars_v42", "_type": "_doc", "_id": "1", "_score": 2.6568544, "_source": { "quote": "These are <em>not</em> the droids you are looking for." } } ]

Decompounding Commonly in German, Scandinavian languages, Finnish, Korean

PUT /decompound_en { "settings": { "number_of_shards": 1, "analysis": { "filter": { "british_decompounder": { "type": "hyphenation_decompounder", "hyphenation_patterns_path": "hyph/en_GB.xml", "word_list": [ "death", "star" ] } }, "analyzer": { "british_decompound": { "type": "custom", "tokenizer": "standard", "filter": [ "lowercase", "british_decompounder" ] } } } } }

GET /decompound_en/_analyze { "analyzer" : "british_decompound", "text" : "deathstar" }

{ } "tokens": [ { "token": "deathstar", "start_offset": 0, "end_offset": 9, "type": "<ALPHANUM>", "position": 0 }, { "token": "death", "start_offset": 0, "end_offset": 9, "type": "<ALPHANUM>", "position": 0 }, { "token": "star", "start_offset": 0, "end_offset": 9, "type": "<ALPHANUM>", "position": 0 } ]

German Dictionaly (LGPL) https://github.com/uschindler/ german-decompounder

PUT /decompound_de { "settings": { "number_of_shards": 1, "analysis": { "filter": { "german_decompounder": { "type": "hyphenation_decompounder", "word_list_path": "dictionary-de.txt", "hyphenation_patterns_path": "hyph/de_DR.xml", "only_longest_match": true, "min_subword_size": 4 }, "german_stemmer": { "type": "stemmer", "language": "light_german" } }, "analyzer": { "german_decompound": { "type": "custom", "tokenizer": "standard", "filter": [ "lowercase", "german_decompounder", "german_normalization", "german_stemmer" ] } } } } }

GET /decompound_de/_analyze { "analyzer" : "german_decompound", "text" : "Todesstern" }

{ } "tokens": [ { "token": "todesst", "start_offset": 0, "end_offset": 10, "type": "<ALPHANUM>", "position": 0 }, { "token": "tod", "start_offset": 0, "end_offset": 10, "type": "<ALPHANUM>", "position": 0 }, { "token": "stern", "start_offset": 0, "end_offset": 10, "type": "<ALPHANUM>", "position": 0 } ]

Without Word Lists https://github.com/jprante/ elasticsearch-analysis-decompound

Performance

Conclusion

Indexing Formatting Tokenize Lowercase, Stop Words, Stemming Synonyms

Scoring Term Frequency Inverse Document Frequency Field-Length Norm Vector Space Model

Advanced Queries Highlighting Suggestions NGrams, Edge Grams Multiple Analyzers

Advanced Queries Reindex & Alias Update by Query Shingles Decompound

There is more...

Trainings https://training.elastic.co

Thank You! Questions? Philipp Krenn PS: Stickers @xeraa