Elasticsearch query_string通配符不考虑长度

Question

I have some records on Elasticsearch that have the same first letters, such as: word, worda, wordab, wordabc, wordabcd.

I am using query_string with a wildcard:

"query": {
  "bool":{
    "must":[
      {
        "query_string":{
          "query":"word*"
        }
      }
    ]
  }
}

All hits have the same score ("_score" : 1.0), therefore the order is arbitrary. Is it possible to have a score considering how much the word actually matches the term? For instance, word matches the term 100%, worda matches the term 80%, and so on.

Answer 1

The reason why you get score 1 for all matched docs is the following - wildcard/prefix query are multi term queries and in order for them to be executed, Elasticsearch needs to do a rewrite (to get actual matched terms)

There are several ways to achieve this, the default one is called constant_score which assigned all constant scores (ones)

There are several different ways to rewrite - some of them will produce non equal scores, but this scoring would be rather rely on TF-IDF distribution of the terms (eg how often worda is happening in the matched document and how many documents in whole index contains worda ). As a first starting way you could try top_terms_1000 , tweaking it later.

Unfortunately, there is no perfect way out-of-the-box to achieve expected behaviour.

One of the possible ways to mimic it is to try adapt Edge NGram tokenizer to produce tokens from the wordabc as following:

w, wo, wor, word, ...

In this case querying could produce more meaningful score. For perfect expected outcome - percent of the match - you would need to create custom query and scoring mechanism