提高匹配度之加权查询
如果想要对查询结果按某个字段排序，可以用 sort 参数，但 sort 参数会忽视文档匹配度，这样可能会导致匹配度较低的文档位于搜索结果前列。
例如，有如下的文档
albums = (
Album(name='钢铁侠', heat=30, type=1),
Album(name='超人：钢铁之躯', heat=60, type=1),
Album(name='超凡蜘蛛侠', heat=85, type=1),
Album(name='神奇猪猪侠历险记', heat=99, type=2),
)
按 heat 字段降序排列，会导致评分最高的文档排在最后，如下
{
"responseHeader":{
"status":0,
"QTime":1,
"params":{
"sort":"heat desc",
"fl":"name,heat,score",
"indent":"on",
"q":"钢铁侠",
"_":"1478050591433",
"wt":"json"}},
"response":{"numFound":4,"start":0,"maxScore":1.0189849,"docs":[
{
"heat":99,
"name":"神奇猪猪侠历险记",
"score":0.11500417},
{
"heat":85,
"name":"超凡蜘蛛侠",
"score":0.1533389},
{
"heat":60,
"name":"超人：钢铁之躯",
"score":0.25425506},
{
"heat":30,
"name":"钢铁侠",
"score":1.0189849}]
}}
可见，简单的按某个字段升序或降序排列并不足以提高匹配度。
考虑这样一个需求：搜索某个新闻事件(例如美国大选邮件门)，其排序应该是综合考虑关键字匹配度和新闻报道的时间，匹配度 100% 但是时间过去很久，也比不上最近发生但匹配度稍差的新闻。
DixMax/eDisMax 提供了 bq 参数，也许可以满足这样的需求
bq
bq 参数提供了一个子查询，满足该子查询的文档其匹配度会提升。如果有多个条件都要加权，可以传递多个 bq 参数，其格式如下
bq=查询1^加权因子1&bq=查询2^加权因子2&...&bq=查询n^加权因子n
先看下不支持 bq 参数的标准查询解析器的结果
{
"responseHeader":{
"status":0,
"QTime":1,
"params":{
"fl":"name,heat,score",
"indent":"on",
"q":"钢铁侠",
"_":"1478050591433",
"wt":"json"}},
"response":{"numFound":4,"start":0,"maxScore":1.0189849,"docs":[
{
"heat":30,
"name":"钢铁侠",
"score":1.0189849},
{
"heat":60,
"name":"超人：钢铁之躯",
"score":0.25425506},
{
"heat":85,
"name":"超凡蜘蛛侠",
"score":0.1533389},
{
"heat":99,
"name":"神奇猪猪侠历险记",
"score":0.11500417}]
}}
对热度较高的文档加权，即 heat>=80
{
"responseHeader":{
"status":0,
"QTime":1,
"params":{
"fl":"name,heat,score",
"bq":"heat:[80 TO *]",
"indent":"on",
"q":"钢铁侠",
"_":"1478050591433",
"wt":"json",
"defType":"edismax"}},
"response":{"numFound":4,"start":0,"maxScore":0.86861277,"docs":[
{
"heat":30,
"name":"钢铁侠",
"score":0.86861277},
{
"heat":85,
"name":"超凡蜘蛛侠",
"score":0.65355295},
{
"heat":99,
"name":"神奇猪猪侠历险记",
"score":0.6208753},
{
"heat":60,
"name":"超人：钢铁之躯",
"score":0.21673451}]
}}
可以看到
钢铁侠因为热度较低未得到加权，但由于匹配度高，结果依然排第一超人：钢铁之躯由于匹配度并无明显优势，又不是高热度未获得加权，从第二名跌到榜尾
再换个条件试一下，这次试试能否将原本排在榜尾的 神奇猪猪侠历险记 往前排，加权条件为 heat=99
{
"responseHeader":{
"status":0,
"QTime":1,
"params":{
"fl":"name,heat,score",
"bq":"heat:99",
"indent":"on",
"q":"钢铁侠",
"_":"1478050591433",
"wt":"json",
"defType":"edismax"}},
"response":{"numFound":4,"start":0,"maxScore":0.86861277,"docs":[
{
"heat":30,
"name":"钢铁侠",
"score":0.86861277},
{
"heat":99,
"name":"神奇猪猪侠历险记",
"score":0.6208753},
{
"heat":60,
"name":"超人：钢铁之躯",
"score":0.21673451},
{
"heat":85,
"name":"超凡蜘蛛侠",
"score":0.13071059}]
}}
嗯，这个结果还不错的，神奇猪猪侠历险记 虽然提前了，但依然没有影响 钢铁侠 的榜首地位。
bq 的加权因子
上面的试验没有指定加权因子，现在试下加权因子的功效，如下设置 bq 参数
bq=heat:99^2
结果
{
"responseHeader": {
"status": 0,
"QTime": 1,
"params": {
"lowercaseOperators": "true",
"bq": "heat:99^2",
"fl": "name,score",
"indent": "true",
"q": "钢铁侠",
"_": "1478066947552",
"stopwords": "true",
"wt": "json",
"defType": "edismax"
}
},
"response": {
"numFound": 4,
"start": 0,
"maxScore": 0.8477584,
"docs": [
{
"name": "神奇猪猪侠历险记",
"score": 0.8477584
},
{
"name": "钢铁侠",
"score": 0.6438422
},
{
"name": "超人：钢铁之躯",
"score": 0.16065022
},
{
"name": "超凡蜘蛛侠",
"score": 0.09688667
}
]
}
}
在条件查询叠加加权因子的双重激励下，神奇猪猪侠历险记 终于超越 钢铁侠 排名第一了
多重 bq
这一次试下多重 bq，即 bq=heat:99&bq=type:2，只有 神奇猪猪侠历险记 同时满足这 2 个条件，能否威胁 钢铁侠 的榜首地位呢？
查询 url
http://dev:8983/solr/tv/select?bq=heat:99&defType=edismax&fl=name,heat,score&indent=on&q=%E9%92%A2%E9%93%81%E4%BE%A0&wt=json&bq=type:2
查询结果
{
"responseHeader":{
"status":0,
"QTime":2,
"params":{
"fl":"name,heat,score",
"bq":["heat:99",
"type:2"],
"indent":"on",
"q":"钢铁侠",
"wt":"json",
"defType":"edismax"}},
"response":{"numFound":4,"start":0,"maxScore":1.0135437,"docs":[
{
"heat":99,
"name":"神奇猪猪侠历险记",
"score":1.0135437},
{
"heat":30,
"name":"钢铁侠",
"score":0.7697503},
{
"heat":60,
"name":"超人：钢铁之躯",
"score":0.19206655},
{
"heat":85,
"name":"超凡蜘蛛侠",
"score":0.11583357}]
}}
啊哈，神奇猪猪侠历险记 再次胜出
结论
bq参数并不影响查询的结果，影响的是结果的排序bq的加权是有限的，默认情况下并不足以影响匹配度相差较大的文档之间的相对排序bq还可以再叠加加权因子，可调整加权程度(加权因子小于 1 可以减轻加权程度)bq可以叠加多个，对文档匹配度的提升是线性增加的bq涉及的字段可以不建索引，但如果未建索引那么必须是docValues=true
注意
斜体表示未验证