Elasticsearch查询结果异常：匹配"rock coll"时两文档得分为何相同？

Great question! Let's walk through exactly why this is happening and what's going on under the hood:

1. The core issue: Your query terms aren't matching what you think #

First, let's break down how Elasticsearch processes your query and the document text with the default standard analyzer:

• Your query "rock coll" gets split into two separate terms: rock and coll.
• For the second employee's about field ("I like to collect rock albums"), the word collect is analyzed to just collect—not coll. The standard analyzer doesn't do partial/prefix matches or root-word reduction automatically, so coll and collect are treated as completely unrelated terms.

Since neither document actually matches the coll term (only rock is present in both), both docs are only being scored on their match to rock.

2. Match query uses OR logic by default #

The match query defaults to OR behavior when given multiple terms. That means any document matching at least one of the query terms (rock or coll) will be returned. Since both docs have rock, they both make the cut.

3. Score calculation depends on matching terms #

The _score is calculated based on factors like term frequency (how often the matched term appears in the document) and inverse document frequency (how rare the term is across your index). Here:

• Each document has exactly one occurrence of rock
• The rarity of rock is identical across your small index (it appears in both docs)

So the score calculation outputs the exact same value for both documents.

How to get the behavior you expect #

If you want coll to match collect and give the second document a higher score, try these options:

• Use a prefix query: Match terms starting with coll with match_phrase_prefix:

  GET /megacorp/employee/_search
  {
    "query": {
      "match_phrase_prefix": {
        "about": "rock coll"
      }
    }
  }
  

• Add fuzziness: Allow Elasticsearch to match slightly different terms (great for typos or partial matches):

  GET /megacorp/employee/_search
  {
    "query": {
      "match": {
        "about": {
          "query": "rock coll",
          "fuzziness": "AUTO"
        }
      }
    }
  }
  

• Use a custom analyzer: An n-gram analyzer can split text into smaller chunks during indexing, making partial-term matches easier. Or the english analyzer (which does stemming) might help with root-word matches, though note coll isn't a standard stem for collect.

内容的提问来源于stack exchange，提问作者Mandroid