Elasticsearch聚合查询多字段设置权重

你的名字 2022-12-19 11:37 342阅读 0赞

### Elasticsearch聚合查询多字段设置权重 ###

*  背景
 *  环境说明
 *  script设置权重
 *  小结

# 背景 #

实际应用中，可能会需要为为doc文档中某个字段的某些特定的值设置权重，影响排序。es提供了比较灵活的script脚本方式来设置权重，影响评分。以下实现是基于聚合查询的，如果非聚合查询的话，也差不多，可以简单参考下。

# 环境说明 #

1.  开发语言：Java
2.  elasticsearch 版本 7.3.1

# script设置权重 #

初始数据

{
                        "id": "6",
                        “title”:"苹果",
                        "catalogId": "01",
                        "catalogName": "水果",
                        "itemType": "01"
                    },
                    {
                        "id": "7",
                        “title”:"香菇",
                        "catalogId": "02",
                        "catalogName": "蔬菜",
                        "itemType": "02"
                    },
                    {
                        "id": "8",
                        “title”:"芹菜",
                        "catalogId": "02",
                        "catalogName": "蔬菜",
                        "itemType": "02"
                    },
                    {
                        "id": "9",
                        “title”:"鸡肉",
                        "catalogId": "03",
                        "catalogName": "肉类",
                        "itemType": "03"
                    }

以下示例使用catalogId聚合，itemType标识某种业务分类，要求查到的数据中相关度差不多情况下itemType为01的在前，其次是03，最后是02.这里查询条件是match\_all  
废话不多说，上代码！

//获取组的总数
            CardinalityAggregationBuilder totalBuilder = AggregationBuilders.cardinality("total").field("catalogId");
            //groupBy分组统计
            TermsAggregationBuilder termsBuilder = AggregationBuilders.terms("groupByKey").field("catalogId").collectMode(Aggregator.SubAggCollectionMode.BREADTH_FIRST);
            //查询分组后数据
            TopHitsAggregationBuilder groupByKey = AggregationBuilders.topHits("groupByKey")
                    .highlighter(highlightBuilder)
                    .fetchSource(true).size(10)
                    .fetchSource(new String[]{"id","catalogId", "title", "catalogName","itemType"}, new String[]{});
            Map<String, Object> map = new HashMap<>();
            map.put("yilei", "01");
            map.put("yileiWeight", 1.2);
            map.put("erlei", "02");
            map.put("erleiWeight", 0.1);
            map.put("otherWeight", 1);
            //分组评分
            Script script = new Script(
                    ScriptType.INLINE,
                    "painless", 
                    "if(doc['itemType'].value==params.yilei) { return params.yileiWeight * _score} else if(doc['itemType'].value==params.erlei) { return params.erleiWeight * _score} else {return params.otherWeight * _score}", map);
            MaxAggregationBuilder topScore = AggregationBuilders.max("top_score").script(script);
    
            AggregatorFactories.Builder factorBuilder = AggregatorFactories.builder();
            factorBuilder .addAggregator(groupByKey);
            factorBuilder .addAggregator(topScore);
            termsBuilder.subAggregations(factorBuilder );
            BucketOrder bucketOrder = BucketOrder.aggregation("top_score", false);
            termsBuilder.order(bucketOrder);
    
            SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
            BoolQueryBuilder boolQueryBuilder = QueryBuilders.boolQuery();
            boolQueryBuilder.must(QueryBuilders.matchAllQuery());
            searchSourceBuilder.aggregation(termsBuilder );
            searchSourceBuilder.aggregation(totalBuilder);

查询后排序效果如下，达到预期

{
                        "id": "6",
                        "catalogId": "01",
                        "catalogName": "水果",
                        "itemType": "01"
                    },
                    {
                        "id": "9",
                        "catalogId": "03",
                        "catalogName": "肉类",
                        "itemType": "03"
                    },
                    {
                        "id": "7",
                        "catalogId": "02",
                        "catalogName": "蔬菜",
                        "itemType": "02"
                    }

根据top\_score倒排

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3hkMTUwNzQ0ODM0Nzg_size_16_color_FFFFFF_t_70_pic_center]

# 小结 #

是不是即灵活又简单呢！还有以下两点也要关注下：

1.  这种是在**聚合后计算得分**，聚合前也可以使用**ScriptFunction**计算得分，但这样性能会差很多，特别在数据体量大的时候；
2.  计算后的得分不要出现有null的(**有null会使得排序失效**),即脚本中要考虑到每一种情况，给出返回；

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3hkMTUwNzQ0ODM0Nzg_size_16_color_FFFFFF_t_70_pic_center]: /images/20221120/7f9ed743cf374cf49d6445222b5cf0fb.png