solr中facet及facet.pivot理解（整合两篇文章保留参考）

Facet['fæsɪt]很难翻译，只能靠例子来理解了。Solr作者Yonik Seeley也给出更为直接的名字：导航（Guided Navigation）、参数化查询（Paramatic Search）。

上面是比较直接的Faceted Search例子，品牌、产品特征、卖家，均是 Facet 。而Apple、Lenovo等品牌，就是 Facet values 或者说 Constraints ，而Facet values所带的统计值就是 Facet count/Constraint count 。

2 、Facet 使用

q = 超级本 
facet = true 
facet.field = 产品特性 
facet.field = 品牌 
facet.field = 卖家

http://…/select?q=超级本&facet=true&wt=json

&facet.field=品牌&facet.field=产品特性&facet.field=卖家

也可以提交查询条件，设置fq(filter query)。

q = 电脑 
facet = true 
fq = 价格:[8000 TO *] 
facet.mincount = 1 // fq将不符合的字段过滤后，会显示count为0 
facet.field = 产品特性 
facet.field = 品牌 
facet.field = 卖家

http://…/select?q=超级本&facet=true&wt=json

&fq=价格:[8000 TO *]&facet.mincount=1

&facet.field=品牌&facet.field=产品特性&facet.field=卖家

"facet_counts": {
"facet_fields": {
  "品牌": [
    "Apple", 4,
    "Lenovo", 39
      …]
  "产品特性": [
    "显卡", 42,
    "酷睿", 38
      …]
 
  …}}

如果用户选择了Apple这个分类，查询条件中需要添加另外一个fq查询条件，并移除Apple所在的facet.field。

http://…/select?q=超级本&facet=true&wt=json

&fq=价格:[8000 TO *]&fq=品牌:Apple&facet.mincount=1

&facet.field= 品牌 &facet.field=产品特性&facet.field=卖家

3 、Facet 参数

facet.prefix  –   限制constaints的前缀

facet.mincount=0 –  限制constants count的最小返回值，默认为0

facet.sort=count –  排序的方式，根据count或者index

facet.offset=0  –   表示在当前排序情况下的偏移，可以做分页

facet.limit=100 –  constraints返回的数目

facet.missing=false –  是否返回没有值的field

facet.date –  Deprecated, use facet.range

facet.query

指定一个查询字符串作为Facet Constraint

facet.query = rank:[* TO 20]

facet.query = rank:[21 TO *]

"facet_counts": {
"facet_fields": {
  "品牌": [
    "Apple", 4,
    "Lenovo", 10
      …]
  "产品特性": [
    "显卡", 11,
    "酷睿", 20
      …]
 
  …}}

facet.range

http://…/select?&facet=true

&facet.range=price

&facet.range.start=5000

&facet.range.end=8000

&facet.range.gap=1000

<result numFound="27" ... />
 ...
 <lst name="facet_counts">
 <lst name="facet_queries">
   <int name="rank:[* TO 20]">2</int>
   <int name="rank:[21 TO *]">15</int>
 </lst>
...

WARNING:  range范围是左闭右开，[start, end)

facet.pivot

这个是Solr 4.0的新特性，pivot和facet一样难理解，还是用例子来讲吧。

Syntax:  facet.pivot=field1,field2,field3...

e.g.  facet.pivot=comment_user, grade

#docs	#docs grade:好	#docs 等级:中	#docs 等级:差
comment_user:1	10	8	1	1
comment_user:2	20	18	2	0
comment_user:3	15	12	2	1
comment_user:4	18	15	2	1

"facet_counts":{
"facet_pivot":{
 "comment_user, grade ":[{
   "field":"comment_user",
   "value":"1",
   "count":10,
   "pivot":[{
     "field":"grade",
     "value":"好",
     "count":8}, {
     "field":"grade",
     "value":"中",
     "count":1}, {
     "field":"grade",
     "value":"差",
     "count":1}]
   }, {
     "field":" comment_user ",
     "value":"2",
     "count":20,
     "pivot":[{
      …

没有pivot机制的话，要做到上面那点可能需要多次查询：

http://...q= comment&fq= grade:好&facet=true&facet.field=comment_user

http://...q=comment&fq=grade:中&facet=true&facet.field=comment_user

http://...q=comment&fq=grade:差&facet=true&facet.field=comment_user

Facet.pivot -  Computes a Matrix of Constraint Counts across multiple Facet Fields. by Yonik Seeley.

上面那个解释很不错，只能理解不能翻译。

 

facet.pivot自己的理解，就是按照多个维度进行分组查询，以下是自己的实战代码，按照newsType，property两个维度统计：

public List<ReportNewsTypeDTO> queryNewsType(
            ReportQuery reportQuery) {    
        HttpSolrServer solrServer = SolrServer.getInstance().getServer();
        SolrQuery sQuery = new SolrQuery();
        List<ReportNewsTypeDTO> list = new ArrayList<ReportNewsTypeDTO>();
        try {
            String para = this.initReportQueryPara(reportQuery, 0);
            sQuery.setFacet(true);
            sQuery.add("facet.pivot", "newsType,property");//根据这两维度来分组查询
            sQuery.setQuery(para);
            QueryResponse response = solrServer.query(sQuery,SolrRequest.METHOD.POST);     
            NamedList<List<PivotField>> namedList = response.getFacetPivot();
            System.out.println(namedList);//底下为啥要这样判断，把这个值打印出来，你就明白了
            if(namedList != null){
                List<PivotField> pivotList = null;
                for(int i=0;i<namedList.size();i++){
                    pivotList = namedList.getVal(i);
                    if(pivotList != null){
                        ReportNewsTypeDTO dto = null;
                        for(PivotField pivot:pivotList){
                            dto = new ReportNewsTypeDTO();
                            dto.setNewsTypeId((Integer)pivot.getValue());
                            dto.setNewsTypeName(News.newsTypeMap.get((Integer)pivot.getValue()));
                            int pos = 0;
                            int neg = 0;
                            List<PivotField> fieldList = pivot.getPivot();
                            if(fieldList != null){
                                for(PivotField field:fieldList){
                                    int proValue = (Integer) field.getValue();
                                    int count = field.getCount();
                                    if(proValue == 1){
                                        pos = count;
                                    }else{
                                        neg = count;
                                    }
                                }
                            }
                            dto.setPositiveCount(pos);
                            dto.setNegativeCount(neg);
                            list.add(dto);
                        }
                    }
                }
            }

            return list;
        } catch (SolrServerException e) {
            log.error("查询solr失败", e);
            e.printStackTrace();
        } finally{
            solrServer.shutdown();
            solrServer = null;
        }
        return list;    
    }

namedList打印结果:
{newsType,property=
[
newsType:8 [4260] [property:1 [3698] null, property:0 [562] null], 
newsType:1 [1507] [property:1 [1389] null, property:0 [118] null], 
newsType:2 [1054] [property:1 [909] null, property:0 [145] null], 
newsType:6 [715] [property:1 [581] null, property:0 [134] null], 
newsType:4 [675] [property:1 [466] null, property:0 [209] null], 
newsType:3 [486] [property:1 [397] null, property:0 [89] null], 
newsType:7 [458] [property:1 [395] null, property:0 [63] null], 
newsType:5 [289] [property:1 [263] null, property:0 [26] null], 
newsType:9 [143] [property:1 [138] null, property:0 [5] null]
]
}
这下应该明白了。写到这里，突然想到一个，所有的分组查询统计，不管是一个维度两个维度都可以使用face.pivot来统计，不错的东东。