How to fuzzy match email or telephone by Elasticse

问题:

I want to make fuzzy match for email or telephone by Elasticsearch. For example:

match all emails end with @gmail.com

or

match all telephone startwith 136.

I know I can use wildcard,

{
 "query": {
    "wildcard" : {
      "email": "*gmail.com"
    }
  }
}

but the performance is very poor. I tried to use regexp:

{"query": {"regexp": {"email": {"value": "*163\.com*"} } } }

But doesn't work.

Is there better way to make it?

curl -XGET localhost:9200/user_data

{
    "user_data": {
        "aliases": {},
        "mappings": {
            "user_data": {
                "properties": {
                    "address": {
                        "type": "string"
                    },
                    "age": {
                        "type": "long"
                    },
                    "comment": {
                        "type": "string"
                    },
                    "created_on": {
                        "type": "date",
                        "format": "dateOptionalTime"
                    },
                    "custom": {
                        "properties": {
                            "key": {
                                "type": "string"
                            },
                            "value": {
                                "type": "string"
                            }
                        }
                    },
                    "gender": {
                        "type": "string"
                    },
                    "name": {
                        "type": "string"
                    },
                    "qq": {
                        "type": "string"
                    },
                    "tel": {
                        "type": "string"
                    },
                    "updated_on": {
                        "type": "date",
                        "format": "dateOptionalTime"
                    },
                }
            }
        },
        "settings": {
            "index": {
                "creation_date": "1458832279465",
                "uuid": "Fbmthc3lR0ya51zCnWidYg",
                "number_of_replicas": "1",
                "number_of_shards": "5",
                "version": {
                    "created": "1070299"
                }
            }
        },
        "warmers": {}
    }
}

the mapping:

{
  "settings": {
    "analysis": {
      "analyzer": {
        "index_phone_analyzer": {
          "type": "custom",
          "char_filter": [ "digit_only" ],
          "tokenizer": "digit_edge_ngram_tokenizer",
          "filter": [ "trim" ]
        },
        "search_phone_analyzer": {
          "type": "custom",
          "char_filter": [ "digit_only" ],
          "tokenizer": "keyword",
          "filter": [ "trim" ]
        },
        "index_email_analyzer": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [ "lowercase", "name_ngram_filter", "trim" ]
        },
        "search_email_analyzer": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [ "lowercase", "trim" ]
        }
      },
      "char_filter": {
        "digit_only": {
          "type": "pattern_replace",
          "pattern": "\\D+",
          "replacement": ""
        }
      },
      "tokenizer": {
        "digit_edge_ngram_tokenizer": {
          "type": "edgeNGram",
          "min_gram": "3",
          "max_gram": "15",
          "token_chars": [ "digit" ]
        }
      },
      "filter": {
        "name_ngram_filter": {
          "type": "ngram",
          "min_gram": "3",
          "max_gram": "20"
        }
      }
    }
  },
  "mappings" : {
    "user_data" : {
      "properties" : {
        "name" : {
          "type" : "string",
          "analyzer" : "ik"
        },
        "age" : {
          "type" : "integer"
        },
        "gender": {
          "type" : "string"
        },
        "qq" : {
          "type" : "string"
        },
        "email" : {
          "type" : "string",
          "analyzer": "index_email_analyzer",
          "search_analyzer": "search_email_analyzer"
        },
        "tel" : {
          "type" : "string",
          "analyzer": "index_phone_analyzer",
          "search_analyzer": "search_phone_analyzer"
        },
        "address" : {
          "type": "string",
          "analyzer" : "ik"
        },
        "comment" : {
          "type" : "string",
          "analyzer" : "ik"
        },
        "created_on" : {
          "type" : "date",
          "format" : "dateOptionalTime"
        },
        "updated_on" : {
          "type" : "date",
          "format" : "dateOptionalTime"
        },
        "custom": {
          "type" : "nested",
          "properties" : {
            "key" : {
              "type" : "string"
            },
            "value" : {
              "type" : "string"
            }
          }
        }
      }
    }
  }
}

回答1:

An easy way to do this is to create a custom analyzer which makes use of the n-gram token filter for emails (=> see below index_email_analyzer and search_email_analyzer + email_url_analyzer for exact email matching) and edge-ngram token filter for phones (=> see below index_phone_analyzer and search_phone_analyzer).

The full index definition is available below.

PUT myindex
{
  "settings": {
    "analysis": {
      "analyzer": {
        "email_url_analyzer": {
          "type": "custom",
          "tokenizer": "uax_url_email",
          "filter": [ "trim" ]
        },
        "index_phone_analyzer": {
          "type": "custom",
          "char_filter": [ "digit_only" ],
          "tokenizer": "digit_edge_ngram_tokenizer",
          "filter": [ "trim" ]
        },
        "search_phone_analyzer": {
          "type": "custom",
          "char_filter": [ "digit_only" ],
          "tokenizer": "keyword",
          "filter": [ "trim" ]
        },
        "index_email_analyzer": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [ "lowercase", "name_ngram_filter", "trim" ]
        },
        "search_email_analyzer": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": [ "lowercase", "trim" ]
        }
      },
      "char_filter": {
        "digit_only": {
          "type": "pattern_replace",
          "pattern": "\\D+",
          "replacement": ""
        }
      },
      "tokenizer": {
        "digit_edge_ngram_tokenizer": {
          "type": "edgeNGram",
          "min_gram": "1",
          "max_gram": "15",
          "token_chars": [ "digit" ]
        }
      },
      "filter": {
        "name_ngram_filter": {
          "type": "ngram",
          "min_gram": "1",
          "max_gram": "20"
        }
      }
    }
  },
  "mappings": {
    "your_type": {
      "properties": {
        "email": {
          "type": "string",
          "analyzer": "index_email_analyzer",
          "search_analyzer": "search_email_analyzer"
        },
        "phone": {
          "type": "string",
          "analyzer": "index_phone_analyzer",
          "search_analyzer": "search_phone_analyzer"
        }
      }
    }
  }
}

Now, let's dissect it one bit after another.

For the phone field, the idea is to index phone values with index_phone_analyzer, which uses an edge-ngram tokenizer in order to index all prefixes of the phone number. So if your phone number is 1362435647, the following tokens will be produced: 1, 13, 136, 1362, 13624, 136243, 1362435, 13624356, 13624356, 136243564, 1362435647.

Then when searching we use another analyzer search_phone_analyzer which will simply take the input number (e.g. 136) and match it against the phone field using a simple match or term query:

POST myindex
{ 
    "query": {
        "term": 
            { "phone": "136" }
    }
}

For the email field, we proceed in a similar way, in that we index the email values with the index_email_analyzer, which uses an ngram token filter, which will produce all possible tokens of varying length (between 1 and 20 chars) that can be taken from the email value. For instance: [e-mail] will be tokenized to j, jo, joh, ..., gmail.com, ..., [e-mail].

Then when searching, we'll use another analyzer called search_email_analyzer which will take the input and try to match it against the indexed tokens.

POST myindex
{ 
    "query": {
        "term": 
            { "email": "@gmail.com" }
    }
}

The email_url_analyzer analyzer is not used in this example but I've included it just in case you need to match on the exact email value.