Advertising expenditures have become the major source of revenue for e-commerce platforms. Providing good advertising experiences for advertisers through reducing their costs of trial and error for discovering the optimal advertising strategies is crucial for the long-term prosperity of online advertising. To achieve this goal, the advertising platform needs to identify the advertisers' marketing objectives, and then recommend the corresponding strategies to fulfill this objective. In this work, we first deploy a prototype of strategy recommender system on Taobao display advertising platform, recommending bid prices and targeted users to advertisers. We further augment this prototype system by directly revealing the advertising performance, and then infer the advertisers' marketing objectives through their adoptions of different recommending advertising performance. We use the techniques from context bandit to jointly learn the advertisers' marketing objectives and the recommending strategies. Online evaluations show that the designed advertising strategy recommender system can optimize the advertisers' advertising performance and increase the platform's revenue. Simulation experiments based on Taobao online bidding data show that the designed contextual bandit algorithm can effectively optimize the strategy adoption rate of advertisers.
翻译:广告支出已成为电子商务平台的主要收入来源。通过降低广告商在发现最佳广告战略方面的试验成本和错误成本,为广告商提供良好的广告经验,对于网上广告的长期繁荣至关重要。为了实现这一目标,广告平台需要确定广告商的营销目标,然后提出相应的战略建议以实现这一目标。在这项工作中,我们首先在道保展示广告平台上部署战略建议系统原型,向广告商推荐出价和用户;我们通过直接披露广告业绩,进一步强化这一原型系统,然后通过采用不同推荐广告业绩的方法,推介广告商的营销目标。我们利用背景强盗技术共同学习广告商的营销目标和建议战略。在线评估显示,设计广告战略建议系统可以优化广告商的广告业绩,增加平台的收入。基于道保在线招标数据的模拟实验显示,设计的背景强盗算法可以有效地优化广告商的战略采纳率。