Elasticsearch：如何实现对 emoji 表情符号进行搜索

Elasticsearch 是一个应用非常广泛的搜索引擎。它可以对文字进行分词，从而实现全文搜索。在实际的使用中，我们会发现有一些文字中包含一些表情符号，比如笑脸，动物等等，那么我们该如何对这些表情符号来进行搜索呢？

🏻 => 🏻, light skin tone, skin tone, type 1–2
🏼 => 🏼, medium-light skin tone, skin tone, type 3
🏽 => 🏽, medium skin tone, skin tone, type 4
🏾 => 🏾, medium-dark skin tone, skin tone, type 5
🏿 => 🏿, dark skin tone, skin tone, type 6
♪ => ♪, eighth, music, note
♭ => ♭, bemolle, flat, music, note
♯ => ♯, dièse, diesis, music, note, sharp
😀 => 😀, face, grin, grinning face
😃 => 😃, face, grinning face with big eyes, mouth, open, smile
😄 => 😄, eye, face, grinning face with smiling eyes, mouth, open, smile
😁 => 😁, beaming face with smiling eyes, eye, face, grin, smile
😆 => 😆, face, grinning squinting face, laugh, mouth, satisfied, smile
😅 => 😅, cold, face, grinning face with sweat, open, smile, sweat
🤣 => 🤣, face, floor, laugh, rofl, rolling, rolling on the floor laughing, rotfl
😂 => 😂, face, face with tears of joy, joy, laugh, tear
🙂 => 🙂, face, slightly smiling face, smile
🙃 => 🙃, face, upside-down
😉 => 😉, face, wink, winking face

🐅 => 🐅, tiger
🐆 => 🐆, leopard
🐴 => 🐴, face, horse
🐎 => 🐎, equestrian, horse, racehorse, racing
🦄 => 🦄, face, unicorn
🦓 => 🦓, stripe, zebra
🦌 => 🦌, deer

在上面，我们可以看到各种各样的 emoji 符号。比如我们想搜索 grin，那么它就把含有 😀 emoji 符号的文档也找出来。在今天的文章中，我们来展示如何实现对 emoji 符号的进行搜索。

安装

如果你还没有对 Elasticsearch 及 Kibana 进行安装的话，请参阅之前的文章 “Elastic：菜鸟上手指南” 进行安装。 另外，我们必须安装 ICU analyzer。关于 ICU analyzer 的安装，请参阅之前的文章 “Elasticsearch：ICU 分词器介绍”。我们在 Elasticsearch 的安装根目录中，打入如下的命令：

./bin/elasticsearch-plugin install analysis-icu

等安装好后，我们需要重新启动 Elasticsearch 让它起作用。运行：

./bin/elasticsearch-plugin list

上面的命令显示：

$ ./bin/elasticsearch-plugin install analysis-icu
-> Installing analysis-icu
-> Downloading analysis-icu from elastic
[=================================================] 100%   
-> Installed analysis-icu
$ ./bin/elasticsearch-plugin list
analysis-icu

安装完 ICU analyzer 后，我们必须重新启动 Elasticsearch。

搜索 emoji 符号

我们先做一个简单的实验：

GET /_analyze
{
  "tokenizer": "icu_tokenizer",
  "text": "I live in 🇨🇳 and I'm 👩‍🚀"
}

上面使用 icu_tokenizer 来对 “I live in 🇨🇳  and I'm 👩‍🚀” 进行分词。 👩‍🚀 表情符号非常独特，因为它是更经典的 👩 和 🚀 表情符号的组合。 中国的国旗也很特别，它是 🇨 和 🇳 的组合。 因此，我们不仅在谈论正确地分割 Unicode 代码点，而且在这里真正地了解了表情符号。

上面的请求的返回结果为：

{
  "tokens" : [
    {
      "token" : "I",
      "start_offset" : 0,
      "end_offset" : 1,
      "type" : "<ALPHANUM>",
      "position" : 0
    },
    {
      "token" : "live",
      "start_offset" : 2,
      "end_offset" : 6,
      "type" : "<ALPHANUM>",
      "position" : 1
    },
    {
      "token" : "in",
      "start_offset" : 7,
      "end_offset" : 9,
      "type" : "<ALPHANUM>",
      "position" : 2
    },
    {
      "token" : """🇨🇳""",
      "start_offset" : 10,
      "end_offset" : 14,
      "type" : "<EMOJI>",
      "position" : 3
    },
    {
      "token" : "and",
      "start_offset" : 16,
      "end_offset" : 19,
      "type" : "<ALPHANUM>",
      "position" : 4
    },
    {
      "token" : "I'm",
      "start_offset" : 20,
      "end_offset" : 23,
      "type" : "<ALPHANUM>",
      "position" : 5
    },
    {
      "token" : """👩‍🚀""",
      "start_offset" : 24,
      "end_offset" : 29,
      "type" : "<EMOJI>",
      "position" : 6
    }
  ]
}

显然 emoji 的符号被正确地分词，并能被搜索。

在实际的使用中，我们可能并不限限于对这些 emoji 的符号的搜索。比如我们想对如下的文档进行搜索：

PUT emoji-capable/_doc/1
{
  "content": "I like 🐅"
}

上面的文档中含有一个 🐅，也就是老虎。针对上面的文档，我们想搜索 tiger 的时候，也能正确地搜索到文档，那么我们该如何去做呢？

在 github 上面，有一个项目叫做 https://github.com/jolicode/emoji-search/。在它的项目中，有一个目录 https://github.com/jolicode/emoji-search/tree/master/synonyms。这里其实就是同义词的目录。我们现在下载其中的一个文件 https://github.com/jolicode/emoji-search/blob/master/synonyms/cldr-emoji-annotation-synonyms-en.txt 到 Elasticsearch 的本地安装目录：

config
├── analysis
│   ├── cldr-emoji-annotation-synonyms-en.txt
│   └── emoticons.txt
├── elasticsearch.yml
...

在我的电脑上：

$ pwd
/Users/liuxg/elastic1/elasticsearch-7.11.0/config
$ tree -L 3
.
├── analysis
│   └── cldr-emoji-annotation-synonyms-en.txt
├── elasticsearch.keystore
├── elasticsearch.yml
├── jvm.options
├── jvm.options.d
├── log4j2.properties
├── role_mapping.yml
├── roles.yml
├── users
└── users_roles

在上面的 cldr-emoji-annotation-synonyms-en.txt 的文件中，它包含了常见 emoji 的符号的同义词。比如：

😀 => 😀, face, grin, grinning face
😃 => 😃, face, grinning face with big eyes, mouth, open, smile
😄 => 😄, eye, face, grinning face with smiling eyes, mouth, open, smile
😁 => 😁, beaming face with smiling eyes, eye, face, grin, smile
😆 => 😆, face, grinning squinting face, laugh, mouth, satisfied, smile
😅 => 😅, cold, face, grinning face with sweat, open, smile, sweat
....

为此，我们来进行如下的实验：

PUT /emoji-capable
{
  "settings": {
    "analysis": {
      "filter": {
        "english_emoji": {
          "type": "synonym",
          "synonyms_path": "analysis/cldr-emoji-annotation-synonyms-en.txt" 
        }
      },
      "analyzer": {
        "english_with_emoji": {
          "tokenizer": "icu_tokenizer",
          "filter": [
            "english_emoji"
          ]
        }
      }
    }
  },
  "mappings": {
    "properties": {
      "content": {
        "type": "text",
        "analyzer": "english_with_emoji"
      }
    }
  }
}

在上面，我们定义了 english_with_emoji 分词器，同时我们在定义 content 字段时也使用相同的分词器 english_with_emoji。我们使用 _analyze API 来进行如下的使用：

GET emoji-capable/_analyze
{
  "analyzer": "english_with_emoji",
  "text": "I like 🐅"
}

上面的命令返回：

{
  "tokens" : [
    {
      "token" : "I",
      "start_offset" : 0,
      "end_offset" : 1,
      "type" : "<ALPHANUM>",
      "position" : 0
    },
    {
      "token" : "like",
      "start_offset" : 2,
      "end_offset" : 6,
      "type" : "<ALPHANUM>",
      "position" : 1
    },
    {
      "token" : """🐅""",
      "start_offset" : 7,
      "end_offset" : 9,
      "type" : "SYNONYM",
      "position" : 2
    },
    {
      "token" : "tiger",
      "start_offset" : 7,
      "end_offset" : 9,
      "type" : "SYNONYM",
      "position" : 2
    }
  ]
}

显然它除了返回 🐅， 也同时返回了 tiger 这样的 token。也就是说我们可以同时搜索这两种，都可以搜索到这个文档。同样地：

GET emoji-capable/_analyze
{
  "analyzer": "english_with_emoji",
  "text": "😀 means happy"
}

它返回：

{
  "tokens" : [
    {
      "token" : """😀""",
      "start_offset" : 0,
      "end_offset" : 2,
      "type" : "SYNONYM",
      "position" : 0
    },
    {
      "token" : "face",
      "start_offset" : 0,
      "end_offset" : 2,
      "type" : "SYNONYM",
      "position" : 0
    },
    {
      "token" : "grin",
      "start_offset" : 0,
      "end_offset" : 2,
      "type" : "SYNONYM",
      "position" : 0
    },
    {
      "token" : "grinning",
      "start_offset" : 0,
      "end_offset" : 2,
      "type" : "SYNONYM",
      "position" : 0
    },
    {
      "token" : "means",
      "start_offset" : 3,
      "end_offset" : 8,
      "type" : "<ALPHANUM>",
      "position" : 1
    },
    {
      "token" : "face",
      "start_offset" : 3,
      "end_offset" : 8,
      "type" : "SYNONYM",
      "position" : 1
    },
    {
      "token" : "happy",
      "start_offset" : 9,
      "end_offset" : 14,
      "type" : "<ALPHANUM>",
      "position" : 2
    }
  ]
}

它表明，如果我们搜索 face, grinning，grin，该文档也会被正确地返回。

现在，我们输入如下的两个文档：

PUT emoji-capable/_doc/1
{
  "content": "I like 🐅"
}

PUT emoji-capable/_doc/2
{
  "content": "😀 means happy"
}

我们对文档进行如下的搜索：

GET emoji-capable/_search
{
  "query": {
    "match": {
      "content": "🐅"
    }
  }
}

或：

GET emoji-capable/_search
{
  "query": {
    "match": {
      "content": "tiger"
    }
  }
}

他们都将返回第一个文档：

{
  "took" : 2,
  "timed_out" : false,
  "_shards" : {
    "total" : 1,
    "successful" : 1,
    "skipped" : 0,
    "failed" : 0
  },
  "hits" : {
    "total" : {
      "value" : 1,
      "relation" : "eq"
    },
    "max_score" : 0.8514803,
    "hits" : [
      {
        "_index" : "emoji-capable",
        "_type" : "_doc",
        "_id" : "1",
        "_score" : 0.8514803,
        "_source" : {
          "content" : """I like 🐅"""
        }
      }
    ]
  }
}

通用地，我们进行如下的搜索：

GET emoji-capable/_search
{
  "query": {
    "match": {
      "content": "😀"
    }
  }
}

或者：

GET emoji-capable/_search
{
  "query": {
    "match": {
      "content": "grin"
    }
  }
}

它们都将返回第二个文档：

{
  "took" : 1,
  "timed_out" : false,
  "_shards" : {
    "total" : 1,
    "successful" : 1,
    "skipped" : 0,
    "failed" : 0
  },
  "hits" : {
    "total" : {
      "value" : 1,
      "relation" : "eq"
    },
    "max_score" : 0.8514803,
    "hits" : [
      {
        "_index" : "emoji-capable",
        "_type" : "_doc",
        "_id" : "2",
        "_score" : 0.8514803,
        "_source" : {
          "content" : """😀 means happy"""
        }
      }
    ]
  }
}