Elasticsearch中Match查询无法匹配手机号前缀的问题排查

Alright, let's break down why your match query isn't working for phone number prefix searches and how to fix this issue step by step.

First, let's recap what you're seeing:

• A match query for "555" on the phoneNumber field returns no results
• A prefix query for "555" works perfectly
• The tokenization result shows the full phone number 5551112233 is treated as a single token of type <NUM>

The core issue here is how Elasticsearch handles the phoneNumber field during indexing and search:

• The match query searches against tokenized values of the field. Since your phone number is being tokenized as one complete string, the query term "555" (which gets tokenized to "555") doesn't match the full-length token.
• The prefix query skips tokenization and matches directly against the raw field value, which is why it works.

Looking at your provided mapping, you didn't specify any special configuration for the phoneNumber field—so it's using the index's default analyzer (likely the standard analyzer here), which treats continuous digits as a single token.

Here are the most effective ways to fix this for phone number partial/prefix search:

1. Use Edge Ngram Analyzer (Recommended) #

You already have an AutoCompleteAnalyzer defined with edge ngram filtering—this is perfect for prefix-based search. You just need to assign it to the phoneNumber field in your mapping.

Update your mapping configuration like this:

{
  "index": "user-clinics",
  "type": "user-clinic",
  "body": {
    "properties": {
      "id": { "type": "long" },
      "userId": { "type": "long" },
      "name": { "type": "text" },
      "phoneNumber": {
        "type": "text",
        "analyzer": "autocomplete_index",
        "search_analyzer": "autocomplete_search"
      }
    }
  }
}

• When indexing, the autocomplete_index analyzer will split 5551112233 into edge ngram tokens like 5, 55, 555, 5551, ..., 5551112233
• When searching with "555", the autocomplete_search analyzer will tokenize the query to "555", which will match the corresponding ngram token from the indexed field.

2. Mark phoneNumber as a Keyword Type (Simple but Limited) #

If you only need exact or prefix matches (no mid-number searches), you can set the phoneNumber field to keyword type. This skips tokenization entirely:

{
  "index": "user-clinics",
  "type": "user-clinic",
  "body": {
    "properties": {
      "id": { "type": "long" },
      "userId": { "type": "long" },
      "name": { "type": "text" },
      "phoneNumber": { "type": "keyword" }
    }
  }
}

With this setup, you can use either the prefix query or a match_phrase_prefix query to find results. Note that this won't support searches for middle segments (like "111" in your example number).

3. Wildcard Query (Not Recommended for Large Datasets) #

If you can't modify the mapping right now, a wildcard query will work—but it's inefficient for large indices because it has to scan many documents:

{"wildcard" : {"phoneNumber": {"value": "555*"}}}

After updating the mapping, you can test the tokenization with the _analyze API to confirm it's working:

POST /user-clinics/_analyze
{
  "analyzer": "autocomplete_index",
  "text": "5551112233"
}

You should see a list of edge ngram tokens starting from the first digit, which means your match query for "555" will now return the expected result.

内容的提问来源于stack exchange，提问作者Raven