如何存储房地产属性

Question

我对Elasticsearch非常陌生。我有一些文件可以具有这样的属性：

1. 浴室内没有
2. 卧室
3. Zip
4. 地址

我想将这些属性存储在一个单独的字段中，这样用户就可以使用"3张床在97778(zip)“进行搜索。

我试过用一个数组字段，例如3张床、2张浴室、97778张床、3张浴室、97778张塞字分析器，这样我就可以限制这类词中的“at”、"at"，但这似乎不是正确的方法，因为第二个文档的得分高于第一个文档。

此外，我有一个同义词分析器，因为如果一个用户搜索与“3BD”，它应该返回"3床位“。

现在我的问题是，存储属性的最佳方法是什么？这是我的一些假文件。

{
    "Beds" : 3,
    "Bath" : 2,
    "Zip" : 97778,
    "Attributes" : ["3 beds","2 baths", "97778"]
},
{
    "Beds" : 7,
    "Bath" : 3,
    "Zip" : 97778,
    "Attributes" : [7 beds,3 baths, 97778]
}

我是否应该将此架构更改为

{
    "Beds" : 7,
    "Bath" : 3,
    "Zip" : 97778,
    "Attributes" : [bed : "7", bath : "3", zip : "97778"]
}

如果是这样的话，那我该如何使用同义词分析器呢？

Answer 1

第一种结构看起来更适合我。我使用Marvel在本地机器上使用这些属性创建了一个简单的索引：

PUT /test
{
  "settings": {
    "analysis": {
      "filter": {
        "my_stop": {
          "type":       "stop",
          "stopwords":  "_english_" 
        },
        "my_possessive_stemmer": {
          "type":       "stemmer",
          "language":   "possessive_english"
        },
        "my_synonym": {
          "type": "synonym",
          "synonyms": [
            "bd => bed",
            "bt, baths, bth => bath"]
        },
        "my_shingle": {
          "type" : "shingle",
          "min_shingle_size": 2,
          "max_shingle_size": 3,
          "output_unigrams": false,
          "output_unigrams_if_no_shingles": true
        }
      },
      "analyzer": {
        "my_english": {
          "tokenizer":  "standard",
          "filter": [
            "my_possessive_stemmer",
            "lowercase",
            "my_stop",
            "my_synonym",
            "kstem",
            "my_shingle"
          ]
        }
      }
    }
  },
  "mappings": {
    "documents": {
      "properties": {
        "Beds": {
          "type": "integer"
        },
        "Baths": {
          "type": "integer"
        },
        "Zip": {
          "type": "integer"
        },
        "Attributes": {
          "type": "string",
          "analyzer": "my_english"
        }
      }
    }
  }
}

这是一个标准的英语分析器(我只排除了词干过滤器，在我看来，它太咄咄逼人了，代之以kstem)，当然还有你的同义词。我还添加了板条过滤器，它产生令牌组合，这是我们正在寻找的！

我增加了你的测试数据。请注意，我已经加倍关键字zip，以防用户希望查找zip 97778或97778 zip，这可能是可能的。

PUT /test/documents/1
{
  "Beds": 3,
  "Bath": 2,
  "Zip": 97778,
  "Attributes": ["3 beds", "2 baths", "zip 97778 zip"]
}

PUT /test/documents/2
{
  "Beds": 7,
  "Bath": 3,
  "Zip": 97778,
  "Attributes": ["7 beds", "3 baths", "zip 97778 zip"]
}

POST /test/documents/3
{
  "Attributes" : ["8310 prairie rose place", "md", "baltimore", "21208", "us", "3 bd", "3 bth", "1 pbh", "1 hbh", "cooktop", "dishwasher", "dryer", "garbage disposer", "ice maker", "microwave", "oven", "oven - double", "refrigerator", "washer", "appliances", "contemporary architecture", "ceiling fan(s)", "colling system", "brick", "basement", "forced air", "heating system", "3 floors", "2 parkings", "garage", "asphalt roof"]
}

POST /test/documents/4
{
  "Attributes" : ["8 winners circle", "md", "owings mills", "21117", "us", "2 bd", "1 bth", "dishwasher", "dryer", "garbage disposer", "microwave", "range", "refrigerator", "washer", "appliances", "traditional architecture", "new traditional architecture", "central a/c", "colling system", "vinyl siding", "heat pump", "heating system", "1 floors", "assigned", "unassigned", "unknown roof"]
}

这是一个简单的匹配查询：

POST /test/documents/_search
{
  "query": {
    "match": {
      "Attributes": {
        "query": "3 beds at 97778(zip)"
      }
    }
  }
}

它根据要求提供所需的数据：

{
  "_index" : "test",
  "_type" : "documents",
  "_id" : "1",
  "_score" : 0.020668881,
  "_source" : {
    "Beds" : 3,
    "Bath" : 2,
    "Zip" : 97778,
    "Attributes" : [
      "3 beds",
      "2 baths",
      "zip 97778 zip"
    ]
  }
},
{
  "_index" : "test",
  "_type" : "documents",
  "_id" : "2",
  "_score" : 0.004767749,
  "_source" : {
    "Beds" : 7,
    "Bath" : 3,
    "Zip" : 97778,
    "Attributes" : [
      "7 beds",
      "3 baths",
      "zip 97778 zip"
    ]
  }
},
{
  "_index" : "test",
  "_type" : "documents",
  "_id" : "3",
  "_score" : 0.0014899216,
  "_source" : {
    "Attributes" : [
      "8310 prairie rose place",
      "md",
      "baltimore",
      "21208",
      "us",
      "3 bd",
      "3 bth",
      "1 pbh",
      "1 hbh",
      "cooktop",
      "dishwasher",
      "dryer",
      "garbage disposer",
      "ice maker",
      "microwave",
      "oven",
      "oven - double",
      "refrigerator",
      "washer",
      "appliances",
      "contemporary architecture",
      "ceiling fan(s)",
      "colling system",
      "brick",
      "basement",
      "forced air",
      "heating system",
      "3 floors",
      "2 parkings",
      "garage",
      "asphalt roof"
    ]
  }
}

现在当我问这个问题时：

POST /test/documents/_search
{
  "query": {
    "match": {
      "Attributes": {
        "query": "2 bd and 1 bth at md"
      }
    }
  }
}

它返回这个结果，这是正确的：

{
  "_index" : "test",
  "_type" : "documents",
  "_id" : "4",
  "_score" : 0.0032357208,
  "_source" : {
    "Attributes" : [
      "8 winners circle",
      "md",
      "owings mills",
      "21117",
      "us",
      "2 bd",
      "1 bth",
      "dishwasher",
      "dryer",
      "garbage disposer",
      "microwave",
      "range",
      "refrigerator",
      "washer",
      "appliances",
      "traditional architecture",
      "new traditional architecture",
      "central a/c",
      "colling system",
      "vinyl siding",
      "heat pump",
      "heating system",
      "1 floors",
      "assigned",
      "unassigned",
      "unknown roof"
    ]
  }
}

您说您的结果总是以1得分，这表明您的查询运行不正确。我可以猜到，这个问题是针对attributes字段而不是Attributes运行的，不幸的是，Elasticsearch非常区分大小写。

从评论中，你说你在使用术语查询 --对文本数据使用它是不对的，因为它总是在寻找准确的术语匹配。总是在搜索文本数据时使用匹配查询。

如果这有帮助的话请告诉我。