In ecommerce platform, users will be less likely to use organic search if sponsored search shows them unexpected advertising items, which will be a hidden cost for the platform. In order to incorporate the hidden cost into auction mechanism which helps create positive growth for the platform, we turn to a reserve price design to decide whether we sell the traffic, as well as build healthy relationships between revenue and user experience. We propose a dynamic reserve price design framework to sell traffic more efficiently with minimal cost of user experience while keeping long term incentives to the advertisers to reveal their valuations truthfully. A distributed algorithm is also proposed to compute the reserve price with billion scale data in the production environment. Experiments with offline evaluations and online AB testing demonstrate that it is a simple and efficient method to be suitably used in industrial production. It has already been fully deployed in the production of Lazada sponsored search.
翻译:在电子商务平台上,如果赞助的搜索显示出意想不到的广告项目,用户就不太可能使用有机搜索,这是平台的隐藏成本。为了将隐藏成本纳入拍卖机制,为平台带来积极增长,我们转向储备价格设计,以确定我们是否出售交易,以及建立收入和用户经验之间的健康关系。我们提议一个动态储备价格设计框架,以用户经验的最低成本更高效地销售交易,同时长期鼓励广告商真实地披露其价值。还提议采用分布式算法,用生产环境中的10亿比例数据计算储备价格。离线评估和在线AB测试实验表明,这是在工业生产中适当使用的一种简单而有效的方法。我们已在生产拉扎达赞助的搜索中充分运用了这一方法。