導語
search_as_you_type 字段類型是一個類似 text 的字段，經過優化，可以爲提供按需輸入完成情況的查詢提供開箱即用的支持。它創建了一系列子字段，這些子字段被分析以索引可被部分與整個索引文本值匹配的查詢有效匹配的術語。支持前綴完成（即，匹配項從輸入的開頭開始）和中綴完成（即，匹配項在輸入中的任意位置）。
將這種類型的字段添加到 mapping 時
PUT my-index-000001{"mappings": {"properties": {"my_field": {"type": "search_as_you_type"}}}}
這將創建以下字段：
|
my_field |
按照 mapping 中的配置進行分析。如果未配置分析器，則使用索引的默認分詞器 |
|
my_field._2gram |
用大小爲 2 的 shingle token filter 分詞器對 ny_field 進行分詞 |
|
my_field._3gram |
用大小爲 3 的 shingle token filter 分詞器對 ny_field 進行分詞 |
|
my_field._index_prefix |
用 edge ngram token filter 包裝 my_field._3gram 的分詞器 |
可以使用 max_shingle_size mapping 參數配置子字段中的 shingles 的大小。默認值爲3，此參數的有效值爲整數值2 - 4（含2和4）。將爲從 2 到max_shingle_size（包括max_shingle_size）的每個 shingle size 創建 shingle 子字段。在構造自己的分析器時，my_field._index_prefix 子字段將始終使用帶有 max_shingle_size 的 shingle 子字段中的分析器。
增加 max_shingle_size 將改善具有更多連續項的查詢的匹配度，但代價是較大的索引大小。默認的 max_shingle_size 通常應該足夠了。
當被索引的文檔具有根字段 my_field 的值時，相同的輸入文本將使用不同的分析鏈自動索引到這些字段中的每個字段中。
上面的描述確實有些拗口，不太容易理解。
我們還是拿一個簡單的例子來展示：
PUT jobs{"mappings": {"properties": {"title": {"type": "search_as_you_type"}}}}
在上面，我們創建了一個叫做 jobs 的索引。在這個索引中，我們定義了一個叫做 search_as_you_type 的字段 title。
我們來使用如下的 _analyze API 來進行展示：
POST jobs/_analyze{"field": "title","text": ["Senior Software Developer"]}
上面的結果顯示：
{"tokens" : [{"token" : "senior","start_offset" : 0,"end_offset" : 6,"type" : "<ALPHANUM>","position" : 0},{"token" : "software","start_offset" : 7,"end_offset" : 15,"type" : "<ALPHANUM>","position" : 1},{"token" : "developer","start_offset" : 16,"end_offset" : 25,"type" : "<ALPHANUM>","position" : 2}]}
這個和我們之前看的沒有什麼兩樣。但是在上面我們也講到，它將生成其它的字段：
title._2gram
title._3gram
title._index_prefix
我們可以使用如下的方法來進行測試：
POST jobs/_analyze{"field": "title._2gram","text": ["Senior Software Developer"]}
上面的結果顯示：
{"tokens" : [{"token" : "senior software","start_offset" : 0,"end_offset" : 15,"type" : "shingle","position" : 0},{"token" : "software developer","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1}]}
也就是說，當我們查詢 senior software 或者 software developer 時，這個文檔將會被搜索到。
同樣地，我們針對 title._3gram 來做分析：
POST jobs/_analyze{"field": "title._3gram","text": ["Senior Software Developer"]}
上面的分析結果爲：
{"tokens" : [{"token" : "senior software developer","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0}]}
也就是說當我們完整地輸入 senior software developer，那麼這個文檔將會被搜索到。
接下來，我們對 title._index_prefix 來進行分析：
POST jobs/_analyze{"field": "title._index_prefix","text": ["Senior Software Developer"]}
這個返回結果是非常之長的：
{"tokens" : [{"token" : "s","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "se","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "sen","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "seni","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senio","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior ","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior s","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior so","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior sof","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior soft","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior softw","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior softwa","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior softwar","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior software","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior software ","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior software d","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior software de","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior software dev","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior software deve","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "senior software developer","start_offset" : 0,"end_offset" : 25,"type" : "shingle","position" : 0},{"token" : "s","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "so","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "sof","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "soft","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "softw","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "softwa","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "softwar","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "software","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "software ","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "software d","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "software de","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "software dev","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "software deve","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "software devel","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "software develo","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "software develop","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "software develope","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "software developer","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "software developer ","start_offset" : 7,"end_offset" : 25,"type" : "shingle","position" : 1},{"token" : "d","start_offset" : 16,"end_offset" : 25,"type" : "shingle","position" : 2},{"token" : "de","start_offset" : 16,"end_offset" : 25,"type" : "shingle","position" : 2},{"token" : "dev","start_offset" : 16,"end_offset" : 25,"type" : "shingle","position" : 2},{"token" : "deve","start_offset" : 16,"end_offset" : 25,"type" : "shingle","position" : 2},{"token" : "devel","start_offset" : 16,"end_offset" : 25,"type" : "shingle","position" : 2},{"token" : "develo","start_offset" : 16,"end_offset" : 25,"type" : "shingle","position" : 2},{"token" : "develop","start_offset" : 16,"end_offset" : 25,"type" : "shingle","position" : 2},{"token" : "develope","start_offset" : 16,"end_offset" : 25,"type" : "shingle","position" : 2},{"token" : "developer","start_offset" : 16,"end_offset" : 25,"type" : "shingle","position" : 2},{"token" : "developer ","start_offset" : 16,"end_offset" : 25,"type" : "shingle","position" : 2},{"token" : "developer ","start_offset" : 16,"end_offset" : 25,"type" : "shingle","position" : 2}]}
首先，我們必須知道的是，它是針對 title._3gram （senior software developer）來進行 dge ngram token filter 操作的。從上面的結果，我們可以看出來，當我們輸入如下的任何一個：
s, se, sen, seni, senio, senior, senior , senior s, senior so, senior sof, senior soft, senior softw, senior softwa, senior softwar, senior software, senior software , senior software d, senior software de, senior software dev, senior software deve, senior software devel, senior software develo, senior software develop, senior software develope, senior software developer
s, so, sof, soft, softw, softwa, softwar, software, software , software d, software de, software dev, software deve, software devel, software develo, software develop, software develope, software developer
d, de, dev, deve, devel, develo, develop, develope, developer
我們的文檔都將被匹配。這樣的好處是我們可以最大限度地匹配我們想要的文檔，但是問題是它將大大增加我們的存儲空間。
下面，我們將使用一個具體的例子來進行展示：
PUT jobs/_bulk?refresh{ "index": {} }{ "title": "Software Developer" }{ "index": {} }{ "title": "Senior Software Developer" }{ "index": {} }{ "title": "Principal Software Developer" }{ "index": {} }{ "title": "Developer Advocate" }{ "index": {} }{ "title": "Developer 🇨🇳" }
在上面，我們創建了5個文檔。
在下面我們將使用 match_phrase_prefix 來進行查詢。
GET jobs/_search{"query": {"match_phrase_prefix": {"title": "developer"}}}
我們將看到5個文檔都將被搜索到。這是因爲它們的分詞裏都含有 developer，當然，我們也甚至可以進行如下的搜索：
GET jobs/_search{"query": {"match_phrase_prefix": {"title": "dev"}}}
它也能搜索所有的5個文檔。
我們進行如下的搜索：
GET jobs/_search{"query": {"match_phrase_prefix": {"title": "software dev"}}}
我們將只搜索到3個文檔：
"hits" : [{"_index" : "jobs","_type" : "_doc","_id" : "MXYGyXgBI6xucLpoKQ9y","_score" : 0.67181337,"_source" : {"title" : "Software Developer"}},{"_index" : "jobs","_type" : "_doc","_id" : "MnYGyXgBI6xucLpoKQ9y","_score" : 0.5679247,"_source" : {"title" : "Software Software Developer"}},{"_index" : "jobs","_type" : "_doc","_id" : "M3YGyXgBI6xucLpoKQ9y","_score" : 0.5679247,"_source" : {"title" : "Principal Software Developer"}}]
如果我們進行如下的搜索：
GET jobs/_search{"query": {"match_phrase_prefix": {"title": "senior dev"}}}
我們將看不到任何的結果。這是因爲在默認的情況下，針對 match_phrase_prefix 的 slop 設置爲 0，也就是說 senior 後面應該馬上接一個以 dev 爲開頭的任何單詞，這樣纔可以進行匹配。
當然，我們可以進行如下的修改：
GET jobs/_search{"query": {"match_phrase_prefix": {"title": {"query": "senior dev","slop": 1}}}}
在上面，我們定義 slop 爲1，表示 senior 和 dev 中間可以插入單詞，這樣也可以進行匹配。上面的搜索結果爲：
"hits" : [{"_index" : "jobs","_type" : "_doc","_id" : "wHYQyXgBI6xucLpoWA-v","_score" : 0.8418889,"_source" : {"title" : "Senior Software Developer"}}]
如果我們想搜索含有 senior 或者以 dev 爲開頭的所有文檔，我們可以這麼搜索：
GET jobs/_search{"query": {"multi_match": {"query": "senior dev","type": "bool_prefix","operator": "or","fields": ["title","title._2gram","title._3gram"]}}}
上面的搜索結果將顯示所有的5個文檔。
我們也可以針對 emoji 符號 🇨🇳 來進行搜索：
GET jobs/_search{"query": {"multi_match": {"query": "🇨🇳","type": "bool_prefix","operator": "and","fields": ["title","title._2gram","title._3gram"]}}}
上面顯示：
"hits" : [{"_index" : "jobs","_type" : "_doc","_id" : "w3YQyXgBI6xucLpoWA-v","_score" : 1.0,"_source" : {"title" : """Developer 🇨🇳"""}}]
總結
針對這種 search_as_you_type，還有一個類型的數據類型 completion。completion suggester 針對速度進行了優化。建議程序使用的數據結構可實現快速查找，但構建成本很高，並且存儲在內存中，同時它不可以做 infix （中綴）查詢。
正文完
作者：劉曉國
本文編輯：喝咖啡的貓
嗨，互動起來吧！
喜歡這篇文章麼？
歡迎留下你想說的，留言 100% 精選哦！
Elastic 社區公衆號長期徵稿，如果您有 Elastic 技術的相關文章，也歡迎投稿至本公衆號，一起進步! 投稿請添加微信：medcl123
招聘信息
Job board
社區招聘欄目是一個新的嘗試，幫助社區的小夥伴找到心儀的職位，也幫助企業找到所需的人才，爲伯樂和千里馬牽線搭橋。有招聘需求的企業和正在求職的社區小夥伴，可以聯繫微信 medcl123 提交招聘需求和發佈個人簡歷信息。
關
注
我
們
Elastic中文社區公衆號 (elastic-cn)
爲您彙集 Elastic 社區的最新動態、精選乾貨文章、精華討論、文檔資料、翻譯與版本發佈等。
喜歡本篇內容就請給我們點個[在看]吧
本文分享自微信公衆號 - Elastic中文社區（elastic-cn）。
如有侵權，請聯繫 [email protected] 刪除。
本文參與“OSC源創計劃”，歡迎正在閱讀的你也加入，一起分享。