弹性搜索更像5.x中的查询分数问题

Question

最近，我们将Elasticsearch版本从2.4更改为5.4。

我们在5.x版本中发现了更多类似于此查询的问题。

下面的查询用于通过文本查找类似的文档

输入查询

POST /test/_search
{
  "size": 10000,
"stored_fields": [
"docid"
],
 "_source": false,
"query": {
"more_like_this": {
"fields": [
    "textcontent"
  ],
  "like": [
    {
      "_index": "test",
      "_type": "object",
      "_id": "AV0c9jvZXF-b5U5aNAWB"
    }
  ],
  "max_query_terms": 5000,
  "min_term_freq": 1,
  "min_doc_freq": 1
}
}
}

弹性搜索2.4的输出

{

"took": 16,
"timed_out": false,
"_shards": {
    "total": 1,
    "successful": 1,
    "failed": 0
},
"hits": {
    "total": 3,
    "max_score": 1.5381224,
    "hits": [
        {
            "_index": "test",
            "_type": "object",
            "_id": "AVzjOOdilllQ-Gyal6Z9",
            "_score": 1.5381224,
            "fields": {
                "docid": [
                    "2"
                ]
            }
        },  {
            "_index": "test",
            "_type": "object",
            "_id": "AVzjOOdilllQ-Gyal63Z",
            "_score": .5381224,
            "fields": {
                "docid": [
                    "3"
                ]
            }
        },  {
            "_index": "test",
            "_type": "object",
            "_id": "AVzjOOdilllQ-Gyal6Z",
            "_score": .381224,
            "fields": {
                "docid": [
                    "4"
                ]
            }
        }

弹性搜索5.4 {的输出

"took": 16,
"timed_out": false,
"_shards": {
    "total": 1,
    "successful": 1,
    "failed": 0
},
"hits": {
    "total": 3,
    "max_score": 1.5381224,
    "hits": [
        {
            "_index": "test",
            "_type": "object",
            "_id": "AVzjOOdilllQ-Gyal6Z9",
            "_score": 168.5381224,
            "fields": {
                "docid": [
                    "2"
                ]
            }
        },  {
            "_index": "test",
            "_type": "object",
            "_id": "AVzjOOdilllQ-Gyal63Z",
            "_score": 164.5381224,
            "fields": {
                "docid": [
                    "3"
                ]
            }
        },  {
            "_index": "test",
            "_type": "object",
            "_id": "AVzjOOdilllQ-Gyal6Z",
            "_score": 132.381224,
            "fields": {
                "docid": [
                    "4"
                ]
            }
        }}

除了文档的得分之外，两个版本的输出都是相同的。5.4版的评分超过2.4分。我们的工作取决于分数，所以如果分数改变了，那对我们来说就是个问题。请提供解决方案？

Answer 1

我得到了解决方案，在5.0版中，他们将默认的相似算法从经典更改为BM25，这就是原因所在。创建索引时，只需将相似类型更改为经典类型即可。如果索引已经存在，那么只需执行以下查询来更新所有索引的设置

PUT /_all/_settings?preserve_existing=true          
{
  "index.similarity.default.type": "classic"
}