wildcardQuery GaussDecayFunctionBuilder 高斯函数比对
1、纯度越高：需求是Formulation中包含的Trade的纯度越高
该配方包含此trade，且国家是140的更靠前，并且包含当前trade的纯度越高越靠前
EsQuery query = new EsQuery();
if(null==size){
query.setPager(0,10);
}else{
query.setPager(0,size);
}
query.includes( "id","tradeName","prd","coverImgId","coverImgSrc","ecoStatus","base64Image" );
//原点（origin）：该字段最理想的值，这个值可以得到满分（1.0） 最佳只包含一个inci
//偏移量（offset）：与原点相差在偏移量之内的值也可以得到满分
//衰减规模（scale）：当值超出了原点到偏移量这段范围，它所得的分数就开始进行衰减了，衰减规模决定了这个分数衰减速度的快慢
//衰减值（decay）：该字段可以被接受的值（默认为 0.5），相当于一个分界点，具体的效果与衰减的模式有关
double origin = 1, offset = 0, scale = 20,decay = 0.5;
//以 e 为底的指数函数
//1、最优先展示只含有当前成分的BASF Product（一对一）
ExponentialDecayFunctionBuilder functionBuilder = ScoreFunctionBuilders.exponentialDecayFunction("inciCount", origin, scale, offset, decay).setWeight(100l);
// 2、优先展示Released Country = Asia Pacific的BASF Product 在字典表中的id:140
BoolQueryBuilder countryBuilder= QueryBuilders.boolQuery()
.must(QueryBuilders.termsQuery("inci_ids_arr",tradeMap.get("id")))
.should(QueryBuilders.matchQuery("releasedCountryId",140).boost(25)); //提高评分
//联合分数
FunctionScoreQueryBuilder scoreQueryBuilder = QueryBuilders.functionScoreQuery(countryBuilder,functionBuilder).boostMode(CombineFunction.SUM);
query.boolQueryBuilder(scoreQueryBuilder);
query.orderByDesc("_score","hitNum","id");
//添加用户角色过滤
EsQueryUtil.addTargetUserAndRole(query);
EsResult result = elasticsearchService.createSearch(EsIndexConstant.es_basf_product,query,Map.class);
resultMaps = (List<Map<String, String>>)result.getData();
for (Map<String, String> map : resultMaps) {
if (tradeMap.get("ecoStatus") != null) {
map.put("base64Image", tradeMapper.getbase64ImageByEco(Integer.valueOf(tradeMap.get("ecoStatus").toString())));
}
}
return resultMaps;
2、相似度越高：Trade中和当前Trade包含的成分，越多匹配越靠前
if (StrUtil.isNotBlank(formulationQuery.getFunctionClaimId()) && !formulationQuery.getFunctionClaimId().equals("0")) {
List<Integer> claimIdList = CollUtil.convertToListFromString(formulationQuery.getFunctionClaimId(), ",");
FunctionScoreQueryBuilder.FilterFunctionBuilder[] filterFunctionBuilders = new FunctionScoreQueryBuilder.FilterFunctionBuilder[claimIdList.size() + 1];
ScoreFunctionBuilder claims = new WeightBuilder();
TermsQueryBuilder claimsQuery = QueryBuilders.termsQuery("functionClaimIdArray", claimIdList);
claims.setWeight(32);
FunctionScoreQueryBuilder.FilterFunctionBuilder claimsBuilder = new FunctionScoreQueryBuilder.FilterFunctionBuilder(claimsQuery, claims);
filterFunctionBuilders[0] = claimsBuilder;
for (Integer claimId : claimIdList) {
ScoreFunctionBuilder scoreFunctionBuilder = new WeightBuilder();
TermQueryBuilder functionClaimIdArray = QueryBuilders.termQuery("functionClaimIdArray", claimId);
scoreFunctionBuilder.setWeight(32f);
FunctionScoreQueryBuilder.FilterFunctionBuilder claim = new FunctionScoreQueryBuilder.FilterFunctionBuilder(functionClaimIdArray, scoreFunctionBuilder);
filterFunctionBuilders[claimIdList.indexOf(claimId) + 1] = claim;
}
boolQueryBuilder.should(QueryBuilders.functionScoreQuery(filterFunctionBuilders)
.scoreMode(FunctionScoreQuery.ScoreMode.SUM)
.boostMode(CombineFunction.REPLACE));
}
3、除了上述两个方式之外 有些算法需要 筛选之后 根据多个条件匹配修改排分 可以使用script
1.第一优先级：子分类相同
2.第三优先级：product form相同
3.第四优先级：product aspect相同
4.第五优先级：SPF相同
5.第六优先级：PA相同
以上条件相同时，按点击数从高到低
“sort”: [
{
”
_script
“: {
“script”: {
“source”: “def total =_score; if(doc[‘countryId’].value ==45) { total+=0.5; } if(doc[‘formId’] ==80) { total+=0.3; } return total”,
“lang”: “painless”
},
“type”: “number”,
“order”: “desc”
}
},
{
“_score”: {
“order”: “desc”
}
},
{
“hitNum”: {
“order”: “desc”
}
},
{
“id”: {
“order”: “desc”
}
}
]
public void fillEsFilterSortScore(EsQuery esQuery, FormulationQuery formulationQuery){
StringBuilder scriptStr = new StringBuilder("def total =_score;");
if (formulationQuery.getFormId() != null && formulationQuery.getFormId() != 0) {
// Create a ScriptSortBuilder for the custom sorting logic
scriptStr.append("if(doc['formId'].value == "+formulationQuery.getFormId()+") {total+=0.5; }");
}
if (formulationQuery.getSpfId() != null && formulationQuery.getSpfId() != 0) {
scriptStr.append("if(doc['spfId'].value == "+formulationQuery.getSpfId()+") {total+=0.3;}");
}
if (formulationQuery.getAspectId() != null && formulationQuery.getAspectId() != 0) {
scriptStr.append("if(doc['aspectId'].value == "+formulationQuery.getAspectId()+") {total+=0.2; }");
}
if (formulationQuery.getPaId() != null && formulationQuery.getPaId() != 0) {
scriptStr.append("if(doc['paId'].value == "+formulationQuery.getPaId()+") {total+=0.1; }");
}
Script script = new Script(scriptStr.toString());
ScriptSortBuilder scriptSortBuilder = SortBuilders.scriptSort(script, ScriptSortBuilder.ScriptSortType.NUMBER);
scriptSortBuilder.order(SortOrder.DESC);
esQuery.sort(scriptSortBuilder);
}
在这个版本的改造过程中发现 第二天这个script竟然失效了，因为有的doc里面formId为空 所以补充修改 如果不为空才进行逻辑判断加分
{
“_script”: {
“script”: {
“source”: “double formV=0,spfV = 0, aspectV=0,paV = 0;if(doc[‘formId’].size()!=0){if(doc[‘formId’].value==80){formV=1.0}}if(doc[‘aspectId’].size()!=0){if(doc[‘aspectId’].value==98){aspectV=0.2}}return formV+aspectV+spfV+paV;”,
“lang”: “painless”
},
“type”: “number”,
“order”: “desc”
}
},