深擎推荐流程代码逻辑

LocalCacheService

ApplicationProperties(配置文件类)

userHistoryMode:String(hazelcast) 配置用户历史的实现方式 hazelcast / redisuserHistoryTopic:String(OUT_HISTORY) 发送历史的topicuserSessionExpireTime:Integer (6) 用户session的过期时间，单位小时userHistoryMaxLimit:Long（1000）每个用户保存的最大的历史数量userInfoQueryMode:String （hazelcast）hazelcast / redisuserHistoryApiCacheExpireTime:Integer (6) 单位小时userStockLimit:Integer（2） 用户关注个股的最大召回量rankModel:RankModeltopN:Integer = 0 （10）自选、持仓、最近交易，主题兴趣等个股数量限制randomM:Integer = 0 （5）topN筛选完后，要在剩余里面随机，保持丰富度recallScoreRadio:Integer （10） 推荐理由得分系数qualityScoreRadio:Integer （50） 质量得分系数

AlikeEngine

obj1:RecConfigHelperobj2:OpServiceobj3:RecommendService

OpService(运营功能Service)

UserHistory

RecommendService

obj:LocalCacheService

RankEngine

obj1:RecConfigHelperobj2:RecommendServiceobj3:HistoryServiceobj4:LocalCacheService

重排

HistoryEngine

obj1:RecConfigHelperobj3:RecommendServiceobj4:HistoryService

推荐引擎

XgbModelHandler

initPosition()    1、记录用户侧整型特征处理          feature list：[\"user_whcd_xl\

FeatureMapsInfo

RankApi（榜单推荐Api）

POST /rank/{laneId}

RerankConfig(重排规则设定)

scatterByItemType:Int 品类打散scatterByContent: lnt 内容打散scatterByReason:Int 推荐理由打散scatterByRatio:Int 比例混合打散scatterByBizPriority:Int 业务优先融合

RecConfigHelper

obj1:RecLanStoreobj2:RecStrategyStoreobj3:RecOpConfigStoreobj4:LocalCacheServiceobj5:RecallProperties

测试(xgb)

POST /api/testing

StrategyItem(推荐策略组合)

strategyId:LongrecalModels:List<ItemModel> 品类策略组合rankModel:Long 唯一排序模型rerankConfig:RerankConfig 重排设定supplment:Supplment 物料补足recallNum:int 召回数量

FeedEngine

obj1:RecConfigHelperobj2:OpServiceobj3:RecommendServiceobj4:HistoryServiceobj5:LocalCacheService

HIstoryApi(用户历史Api)

POST /history/{laneId}

AlikeApi(相关推荐Api)

POST /alike/{laneId}

XgboostRank

featureMapsInfo:FeatureMapsInforecUserInfoStore:RecModelStorerecModelStore:RecModelStoreCATEGORY_RANK:String = \"RANK\"XGB_MODEL_CODE:String = \"XgboostRank\

ScatterUtill

recallItemTypeScatter(List<RecallItem> items):List<RecallItem>:根据品类打散     逐个加入打散结果列表，每次加入时不和上一个品类相同     如果打散后数量不满足，则补齐recallReasonScatter(List<RecallItrem> items):List<RecallItem>根据推荐理由打散      逐个加入打散结果列表，每次加入时不和上一个推荐理由相同      如果打散后数量不满足，则补齐recallItemTypeSequenceScatter(List<RecallItrem> items，List<String> types):List<RecallItem>强制类型轮询      按照指定的品类顺序，依次打散出品          

ReRankItem（重排优先级）

itemType:String 品类reasonCode:String 推荐理由percentage:int 占比pripority:int 优先级

StrategyGroup

planId:计划ID\tstrategyId:策略ID\ttrafficId:流量组ID\tactiveTraffic:是否活跃组\tstrategies:List<StrategyItem>子策略组合\treturnNum:返回数量\treRankSetting:重排设定

TopicApi(主题推荐)

POST /topic/{laneId}

RecEngineApp

TopEngine

obj1:RecConfigHelperobj2:OpServiceobj3:RecommendServiceobj4:LocalCacheService

feedApi(信息流Api)

POST /feed/{laneId}

TopicEngine

obj1:RecConfigHelperobj2:OpServiceobj3:RecommendServiceobj4:HistoryService

HistoryService

ALL_INACTIVE_USER:String = \"ALL_INACTIVE_USER\"userHistory:UserHistory

TopApi(置顶推荐)

POST /top/{laneId}