lucene - Boost 在 Elasticsearch 上不起作用

Question

我是 Elasticsearch 的新手，对 _score 的计算方式感到困惑。我试图通过阅读网上的一些论坛条目( here 和 here )来了解发生了什么，但仍然有一些疑问，无法完全解决我的问题。

瞄准

给定一些包含字段 title 和 content 的文档，查找与 query 匹配的文档，并增强 title 上的匹配 字段。

数据

PUT /sample/myType/1
{
    "title": "Blabbertalk here",
    "content": "Foobar here"
}

PUT /sample/myType/2
{
    "title": "Foobar here",
    "content": "Blabbertalk here"
}

查询

GET /sample/myType/_search
{
    "query": {
        "bool" : {
            "should": [
                {
                    "match" : {
                        "title" : {
                            "query": "Blabbertalk",
                            "fuzziness": 0.7,
                            "boost": 2
                        }
                    }
                },
                {
                    "match" : {
                        "content" : {
                            "query": "Blabbertalk",
                            "fuzziness": 0.7,
                            "boost": 1
                        }
                    }
                }
            ],
            "minimum_number_should_match": 1
        }
    }
}

结果

{
   "took": 21,
   "timed_out": false,
   "_shards": {
      "total": 5,
      "successful": 5,
      "failed": 0
   },
   "hits": {
      "total": 2,
      "max_score": 0.095891505,
      "hits": [
         {
            "_index": "sample",
            "_type": "myType",
            "_id": "1",
            "_score": 0.095891505,
            "_source": {
               "title": "Blabbertalk here",
               "content": "Foobar here"
            }
         },
         {
            "_index": "sample",
            "_type": "myType",
            "_id": "2",
            "_score": 0.095891505,
            "_source": {
               "title": "Foobar here",
               "content": "Blabbertalk here"
            }
         }
      ]
   }
}

问题

虽然这会在顶部返回正确的结果，但在另一种情况下却不会(这促使我问这个问题)。尽管如此，这个样本也有同样让我困惑的问题。

1. 这两个文档的得分非常接近(事实上，太接近了)。我的期望是文档 1 的得分应该更高(大约翻倍)，因为匹配发生在 title 字段
2. 使用 ?explain 参数运行相同的查询会发现 boost 未应用于 _score 计算(见下文)。看来 Elasticsearch 确实能够识别增强因子(如以下行所示:"description": "weight(title:blabbertalk^2.0 in 0) [PerFieldSimilarity], result of:")，但是，进一步查看细节会发现分数计算中实际上没有考虑任何提升因素。
3. 对于上述问题，我对title:blabbertalk^2.0 in 0表示怀疑。 in 0 到底是什么意思？我很确定这并不意味着“在 0 个文档中匹配”。 0 是否会抵消提升？如果是这样，有办法解决这个问题吗？

最后但并非最不重要的一点是，我意识到这可能不是 Elasticsearch 的问题，因为我相信 Elasticsearch 将这个评分和搜索任务委托(delegate)给后端的 Lucene。不过，我对 Lucene 也不太熟悉，所以如果有人能对此有所了解，我将非常感激。

非常感谢您花时间阅读这么长的问题并帮助我。非常感谢，非常感谢!

使用解释参数运行相同的查询

{
   "took": 40,
   "timed_out": false,
   "_shards": {
      "total": 5,
      "successful": 5,
      "failed": 0
   },
   "hits": {
      "total": 2,
      "max_score": 0.095891505,
      "hits": [
         {
            "_shard": 2,
            "_node": "NaOynONhQWSvUmH77e_L9w",
            "_index": "sample",
            "_type": "myType",
            "_id": "1",
            "_score": 0.095891505,
            "_source": {
               "title": "Blabbertalk here",
               "content": "Foobar here"
            },
            "_explanation": {
               "value": 0.095891505,
               "description": "product of:",
               "details": [
                  {
                     "value": 0.19178301,
                     "description": "sum of:",
                     "details": [
                        {
                           "value": 0.19178301,
                           "description": "weight(title:blabbertalk^2.0 in 0) [PerFieldSimilarity], result of:",
                           "details": [
                              {
                                 "value": 0.19178301,
                                 "description": "fieldWeight in 0, product of:",
                                 "details": [
                                    {
                                       "value": 1,
                                       "description": "tf(freq=1.0), with freq of:",
                                       "details": [
                                          {
                                             "value": 1,
                                             "description": "termFreq=1.0"
                                          }
                                       ]
                                    },
                                    {
                                       "value": 0.30685282,
                                       "description": "idf(docFreq=1, maxDocs=1)"
                                    },
                                    {
                                       "value": 0.625,
                                       "description": "fieldNorm(doc=0)"
                                    }
                                 ]
                              }
                           ]
                        }
                     ]
                  },
                  {
                     "value": 0.5,
                     "description": "coord(1/2)"
                  }
               ]
            }
         },
         {
            "_shard": 3,
            "_node": "NaOynONhQWSvUmH77e_L9w",
            "_index": "sample",
            "_type": "myType",
            "_id": "2",
            "_score": 0.095891505,
            "_source": {
               "title": "Foobar here",
               "content": "Blabbertalk here"
            },
            "_explanation": {
               "value": 0.095891505,
               "description": "product of:",
               "details": [
                  {
                     "value": 0.19178301,
                     "description": "sum of:",
                     "details": [
                        {
                           "value": 0.19178301,
                           "description": "weight(content:blabbertalk in 0) [PerFieldSimilarity], result of:",
                           "details": [
                              {
                                 "value": 0.19178301,
                                 "description": "fieldWeight in 0, product of:",
                                 "details": [
                                    {
                                       "value": 1,
                                       "description": "tf(freq=1.0), with freq of:",
                                       "details": [
                                          {
                                             "value": 1,
                                             "description": "termFreq=1.0"
                                          }
                                       ]
                                    },
                                    {
                                       "value": 0.30685282,
                                       "description": "idf(docFreq=1, maxDocs=1)"
                                    },
                                    {
                                       "value": 0.625,
                                       "description": "fieldNorm(doc=0)"
                                    }
                                 ]
                              }
                           ]
                        }
                     ]
                  },
                  {
                     "value": 0.5,
                     "description": "coord(1/2)"
                  }
               ]
            }
         }
      ]
   }
}

Answer 1

我认为 boost 参数应该放在匹配查询本身中，而不是放在字段中，例如:

{
    "query": {
        "bool" : {
            "should": [
                {
                    "match" : {
                        "title" : {
                            "query": "Blabbertalk",
                            "fuzziness": 0.7
                        },
                        "boost": 2
                    }
                },
                {
                    "match" : {
                        "content" : {
                            "query": "Blabbertalk",
                            "fuzziness": 0.7
                        },
                        "boost": 1
                    }
                }
            ],
            "minimum_number_should_match": 1
        }
    }
}

虽然你可以用 multi_match 来简化相反:

{
    "multi_match" : {
        "query":      "Blabbertalk",
        "type":       "most_fields",
        "fields":     [ "title^2", "content" ],
        "fuzziness":  0.7
    }
}