solr debug模式 查看相似度的计算规则

Lucene算法公式如下

score(q,d) = coord(q,d) · queryNorm(q) · ∑ ( tf(t in d) · idf(t)2 · t.getBoost() · norm(t,d) )

• tf(t in d ), = frequency½
• idf(t) = 1 +log(文档总数/(包含t的文档数+1))
• coord(q,d) 评分因子,。越多的查询项在一个文档中，说明些文档的匹配程序越高，比如说，查询"A B C",那么同时包含A/B/C3个词的文档 是3分，只包含A/B的文档是2分，coord可以在query中关掉的
• queryNorm(q)查询的标准查询，使不同查询之间可以比较
• t.getBoost() 和 norm(t,d) 都是提供的可编程接口，可以调整 field/文档/query项 的权重
• queryNorm(q) = 1 / (sumOfSquaredWeights )

  sumOfSquaredWeights = q.getBoost()² • ∑ ( idf(t) • t.getBoost() )² 

• norm(t,d) = d.getBoost() • lengthNorm(f) • f.getBoost() 

• lengthNorm(field) = (1.0 / Math.sqrt(numTerms))：一个域中包含的Term总数越多，也即文档越长，此值越小，文档越短，此值越大。

各种编程插口显得很麻烦，可以不使用，所以我们可以把Lucence的算分公式进行简化

score(q,d) = coord(q,d) · ∑ ( tf(t in d) · idf(t)2 )

"id": "30",
        "sName": "万科 绿地 万科 徐 徐 徐 万科 绿地",
        "text_search": [ "万科 绿地 万科 徐 徐 徐 万科 绿地", "万科 绿地 万科 徐 徐 徐 万科 绿地","万科 绿地 万科 徐 徐 徐 万科 绿地","万科海洋公园","万科海洋公园", "万科海洋绿地公园" ],
        "sAlias": "万科海洋公园",
        "sAddress": "万科海洋绿地公园",
        "_version_": 1554657424083779600 },

 debug": {
    "rawquerystring": "sName:徐",
    "querystring": "sName:徐",
    "parsedquery": "(+sName:徐 FunctionQuery(product(norm(sName),tf(sName,徐),idf(sName,徐))))/no_coord",
    "parsedquery_toString": "+sName:徐 product(norm(sName),tf(sName,徐),idf(sName,徐))",

"explain": {
      "28": "\n1.2036215 = (MATCH) sum of:\n  0.77542824 = (MATCH) weight(sName:徐 in 0) [DefaultSimilarity], result of:\n    0.77542824 = score(doc=0,freq=5.0), product of:\n      0.87540054 = queryWeight, product of:\n        1.8109303 = idf(docFreq=3, maxDocs=9)\n        0.48339826 = queryNorm\n      0.8857982 = fieldWeight in 0, product of:\n        2.236068 = tf(freq=5.0), with freq of:\n          5.0 = termFreq=5.0\n        1.8109303 = idf(docFreq=3, maxDocs=9)\n        0.21875 = fieldNorm(doc=0)\n  0.4281933 = (MATCH) FunctionQuery(product(norm(sName),tf(sName,徐),idf(sName,徐))), product of:\n    0.8857982 = product(norm(sName)=0.21875,tf(sName,徐)=2.236068,idf(sName,徐)=1.8109302520751953)\n    1.0 = boost\n    0.48339826 = queryNorm\n",
      "29": "\n0.761237 = (MATCH) sum of:\n  0.49042383 = (MATCH) weight(sName:徐 in 0) [DefaultSimilarity], result of:\n    0.49042383 = score(doc=0,freq=2.0), product of:\n      0.87540054 = queryWeight, product of:\n        1.8109303 = idf(docFreq=3, maxDocs=9)\n        0.48339826 = queryNorm\n      0.56022793 = fieldWeight in 0, product of:\n        1.4142135 = tf(freq=2.0), with freq of:\n          2.0 = termFreq=2.0\n        1.8109303 = idf(docFreq=3, maxDocs=9)\n        0.21875 = fieldNorm(doc=0)\n  0.27081323 = (MATCH) FunctionQuery(product(norm(sName),tf(sName,徐),idf(sName,徐))), product of:\n    0.560228 = product(norm(sName)=0.21875,tf(sName,徐)=1.4142135,idf(sName,徐)=1.8109302520751953)\n    1.0 = boost\n    0.48339826 = queryNorm\n",

      "30": "\n
  0.9323212 = (MATCH) sum of:\n
  
   0.6006441 = (MATCH) weight(sName:徐 in 0) [DefaultSimilarity], result of:\n  

  //1.
   0.6006441 = score(doc=0,freq=3.0),

   product of:\n  

 // 1.1. 以下为 coord(q,d) · queryNorm(q)     得分
   0.87540054 = queryWeight,

          product of:\n        1.8109303 = idf(docFreq=3, maxDocs=9)\n        0.48339826 = queryNorm\n  

//1.2.以下为bf函数的得分 ( tf(t in d) · idf(t)2 · t.getBoost() · norm(t,d) )
   0.6861363 = fieldWeight in 0, product of:\n       
        1.7320508 = tf(freq=3.0), with freq of:\n          3.0 = termFreq=3.0\n       
        1.8109303 = idf(docFreq=3, maxDocs=9)\n       
        0.21875 = fieldNorm(doc=0)\n 
  

//2.以下为bf函数的得分 ( tf(t in d) · idf(t)2 · t.getBoost() · norm(t,d) )*queryNorm
   0.3316771 = (MATCH) FunctionQuery(product(norm(sName),tf(sName,徐),idf(sName,徐))),
   product of:\n   
   0.6861363 = product(

       norm(sName)=0.21875,

       tf(sName,徐)=1.7320508,

       idf(sName,徐)=1.8109302520751953)\n   

       1.0 = boost\n   

      0.48339826 = queryNorm\n"

0.48339826 = (MATCH) FunctionQuery(product(const(1))), product of:\n    1.0 = product(const(1))\n    1.0 = boost\n    0.48339826 = queryNorm\n"