lucene 评分机制研究(debug信息)

 

lucene评分公式

文档结果：

{
"id": 4111005,
"bookType": 2,
"title": "故宫",
"userId": "xxx",
"userName": "xxx",
"routeDays": 1,
"startTime": 1310313600000,
"cTime": 1407243889812,
"uTime": 1407244293220,
"avgPrice": 0,
"actorType": 0,
"tripType": 0,
"platform": 0,
"quality": 5365590,
"travelMonths": 64,
"sourceId": -1,
"labelName": "",
"imageUrl": "xxx",
"travelRouteTabbed": "故宫",
"travelRoute": "故宫",
"destCitiesTabbed": "北京",
"destCitiesGrouped": "北京",
"destCities": "北京",
"destProvinces": "",
"destNations": "中国",
"destNationsText": "中国",
"destContinents": "亚洲",
"destContinentsText": "亚洲",
"poiIdTabbed": "710603",
"distIds": "299914\t300667",
"provinceName": "",
"nationName": "",
"continentName": "",
"disallowed": "",
"viewCount": 36,
"quoteCount": 0,
"downloadCount": 0,
"ugcType": 2,
"ugcTime": 1407244293220,
"destWeights": "{\"中国\":1.0,\"北京\":1.0,\"china\":1.0}",
"fingerPrint": "153a926f3b27c63889b5dc3e9b69f9ba",
"_version_": 1497957597913809000,
"score": 145735744,
"payload": "score:145735744,matchscore:14.573575,destscore:0.0"
}

得分debug信息，DefaultSimilarity

1 '4111005'=>'//ID号
 2 14.557129 = (MATCH) sum of://最后得分 = 11.167477（相似性评分） + 3.3896518（boost query打分）
 3   11.167477 = (MATCH) max of://相似性评分(max of ：11.167477（命中标题得分），0.40179574（命中途径得分）取其中最大的？)
    // DisjunctionMaxScore源码中对于score的计算为 scoreMax + (scoreSum - scoreMin) * tieBreakerMultipliter
    // tieBreadkerMultipiler传值为0,所以最后就只取最大的匹配值

###################################### 命中标题部分 #####################################
 4     11.167477 = (MATCH) weight(title:故宫^3.0 in 212390) [DefaultSimilarity], result of://命中标题得分明细
 5       11.167477 = score(doc=212390,freq=1.0 = termFreq=1.0)// queryWeight * fieldWeight
 6 , product of:
      //score = queryWeight * fieldWeight = boost * idf * queryNorm * tf * idf * fieldNorm
      //                          = tf * idf^2 * boost * fieldNorm * queryNorm
　　　 //score + score = (tf * idf^2 * boost * fieldNorm + tf * idf^2 *boost *fieldNorm) * queryNorm 与评分公式一致
 7         0.99955523 = queryWeight, product of://queryWeight 为什么不等于boost * idf ？压缩误差？
 8           3.0 = boost //qf=title^3  相当于termQuery.setBoost(3.0)  公式：boost(t.field in d)
 9           11.172446 = idf(docFreq=44, maxDocs=1177741)//逆向文本频率idf（docFreq：该term出现的doc数量44，maxDocs：所有doc数量1177741）
10           0.029822035 = queryNorm //查询权重，不影响排序，多个查询比较使用 公式queryNorm(q)部分
11         11.172446 = fieldWeight in 212390, product of://fieldWeight
12           1.0 = tf(freq=1.0), with freq of: //该term的词频,对应公式tf(t in d)部分
13             1.0 = termFreq=1.0 //该term在本doc中出现的次数
14           11.172446 = idf(docFreq=44, maxDocs=1177741) //逆向文本频率idf(t)部分
15           1.0 = fieldNorm(doc=212390)//公式中lengthNorm部分=state.getboost()*(1/maths.sqrt(numTerms)) numTerms为对应field的term数量 state.getboost()创建索引指定field权重，例中title就是故宫，分词切成一个term，故numTerms=1

#################################### 命中途径部分 #######################################
16     0.40179574 = (MATCH) weight(travelRouteTabbed:故宫^0.5 in 212390) [DefaultSimilarity], result of://命中途径得分
17       0.40179574 = score(doc=212390,freq=1.0 = termFreq=1.0)
18　, product of:
19         0.07740273 = queryWeight, product of:
20           0.5 = boost //qf=travelRouteTabbed:故宫^0.5
21           5.190976 = idf(docFreq=17820, maxDocs=1177741) //逆向文本频率idf（docFreq：该term出现的doc数量17820，maxDocs：所有doc数量1177741）
22           0.029822035 = queryNorm //查询权重，不影响排序，多个查询比较使用 公式queryNorm(q)部分
23         5.190976 = fieldWeight in 212390, product of:
24           1.0 = tf(freq=1.0), with freq of:
25             1.0 = termFreq=1.0
26           5.190976 = idf(docFreq=17820, maxDocs=1177741)
27           1.0 = fieldNorm(doc=212390)

##################################### boost funtion部分 #################################
// 就debug信息来看 没有打印出公式中的coord部分，而是使用lucene原有的公式加上了bf的分数，bf最后的得分依照product of来看应该是 
// sum * boost * queryNorm来得出，那么bf所需要定制的分值就需要考虑乘以queryNorm之后的分数再去影响lucene自身的score，是不是把queryNorm设置为1比较好
// 因为queryNorm对排序的结果没有影响，最终的结果只会被所有命中term的累加和分数再加上bf分数影响，那么coord去哪里呢？
28   3.3896518 = (MATCH) FunctionQuery(sum(rord(routeDays),2000000.0/(1.0*float(rord(quality))+20000.0))), product of://boost query打分）
29     113.66266 = sum(rord(routeDays)=76,2000000.0/(1.0*float(rord(quality)=33103)+20000.0))//bf值的设定
30     1.0 = boost//默认权重是1
31     0.029822035 = queryNorm'