带有 Searckiq 不正确评分的弹性搜索

Question

我有一个模型 Grave，我想在其中查询名称，查询为“Krogness”

我注意到一件事，返回的结果具有相同的分数。 如文档中所述，我将分片数量减少到 1（不走运）。 我什至禁用了拼写错误（运气不好）

下面附上详细的映射和说明。

型号

searchkick word_middle: %i[name plot_number], callbacks: :async, settings: {number_of_shards: 1}

模型 Searchkick 选项

{:word_middle=>[:name, :plot_number],
:callbacks=>:async,
:settings=>{:number_of_shards=>1},
:_type=>
#<Proc:0x00007f53349fcf98@/home/akshayparekh/.rbenv/versions/2.6.0/lib/ruby/gems/2.6.0/bundler/gems/searchkick-bcd6edc872f2/lib/searchkick/model.rb:17 (lambda)>,
:class_name=>"Grave"}

搜索选项

{:fields=>["name^10", "plot_number"],
:where=>{:grave_status=>[:accepted, :claimed], :hidden=>false},
:operator=>:or,
:page=>"1",
:per_page=>30,
:misspellings=>false,
:debug=>true}

查询

curl http://localhost:9200/graves_development/_search?pretty -H 'Content-Type: application/json' -d '{"query":{"bool":{"must":{"bool":{"should":[{"dis_max":{"queries":[{"match":{"name.analyzed":{"query":"Krogness","boost":100.0,"operator":"or","analyzer":"searchkick_search"}}},{"match":{"name.analyzed":{"query":"Krogness","boost":100.0,"operator":"or","analyzer":"searchkick_search2"}}}]}},{"dis_max":{"queries":[{"match":{"plot_number.analyzed":{"query":"Krogness","boost":10,"operator":"or","analyzer":"searchkick_search"}}},{"match":{"plot_number.analyzed":{"query":"Krogness","boost":10,"operator":"or","analyzer":"searchkick_search2"}}}]}}]}},"filter":[{"terms":{"grave_status":["accepted","claimed"]}},{"term":{"hidden":{"value":false}}}]}},"timeout":"11s","_source":false,"size":30}'

结果

{
"took": 2,
"timed_out": false,
"_shards": {
"total": 1,
"successful": 1,
"skipped": 0,
"failed": 0
},
"hits": {
"total": {
"value": 2,
"relation": "eq"
},
"max_score": 1275.969,
"hits": [
{
"_index": "graves_development_20210926173013383",
"_type": "_doc",
"_id": "264729",
"_score": 1275.969
},
{
"_index": "graves_development_20210926173013383",
"_type": "_doc",
"_id": "264730",
"_score": 1275.969
}
]
}
}

我有 2 条记录可以重现此问题。

Test Krogness
Test Krog

当我搜索 Krogness 时，它们的分数相同

Searchkick 版本：4.6.1 Elasticsearch 版本：7.14.1

非常感谢任何帮助

Answer 1

通常情况下，名称​​不 是有意义的（并且不应完全返回“Krog”）。如果你真的需要它们，你应该在你的 ES 映射中有这样的子字段：

"name": {
  "type": "text",
  "fields": {
    "analyzed": {"type": "text", "analyzer": "searchkick_word_middle_index"},
    "lowercase": {"type": "text", "analyzer": "standard"},
    "raw": {"type": "keyword"},
  }
}

然后在搜索阶段，您将拥有：

["name.raw^100", "name.lowercase^50", "name.analyzed^100", ...]

这样，“更精确”的匹配得分会更高。

要让 searchkick 为你应用这些映射，你需要这样的东西：

searchkick merge_mappings: true, mappings: {{
  properties: {
    name: {
      type: "text"
      fields: {...}
    }
  }
}