一种基于异步协同调度的互联网推荐引擎架构设计

Design of Internet Recommendation Engine Architecture Based on Asyn-chronous Cooperative Scheduling

作为互联网信息分发最核心业务,搜索和推荐引擎遵从用户体验优先,以实时响应用户需求构建在线架构,做到秒级响应需求,然而这一设计造就的“召回-粗排-精排”三级漏斗也引入了明显的漏斗损失。文章从推荐引擎差异化场景出发,提出一种基于异步协同调度的新型架构设计,突破性能约束,在用户体验无损前提下做到漏斗精简,显著拉高推荐引擎效率天花板。

As the most important information distribution business of Internet,the search and recommendation engine follows the user experience priority and builds an online architecture to respond to user needs in real time,which can respond at the second level.However,the threc-level funncl of"Recall-Coarse Ranking-Precise Ranking"which created under performance con-straints also causes noticeable funnel losses.Based on the differentiated scenario of recommenda-tion cngine,this paper proposes a new architecture design based on Asynchronous Cooperative Scheduling(ACS),which breaks through performance constraints and achieves funnel simplify without compromising user experience,and significantly raises the efficiency ceiling of recom-mendation engine.