在推荐系统和广告平台上,内容定向推广模块需要尽可能将商品、内容或者广告传递到潜在的对内容感兴趣的用户面前。扩充候选集技术(Look-alike建模)需要基于一个受众种子集合识别出更多的相似潜在用户,从而进行更有针对性的内容投放。然而,look-alike建模通常面临两个挑战:(1)一个系统每天可能需要处理成百上千个不同种类的内容定向推广实例(例如体育、政治、社会等不同领域的内容定向推广)。因此,我们很难构建一个泛化的方法,同时在所有内容领域中扩充高质量的受众候选集。(2)一个内容定向推广任务的受众种子集合可能非常小,而一个基于有限种子用户的定制化模型往往会产生严重的过拟合。怎么解决以上的挑战呢? AI 科技评论今天为大家介绍一篇被KDD-2021收录的论文《Learning to Expand Audience via Meta Hybrid Experts and Critics for Recommendation and Advertising》,论文作者来自中科院计算所、腾讯微信看一看、北航。
论文链接:https://arxiv.org/abs/2105.14688为了解决以上的挑战,这篇论文提出了一种新的两阶段框架Meta Hybrid Experts and Critics (MetaHeac)。
[1] Kanagal B, Ahmed A, Pandey S, et al. Focused matrix factorization for audience selection in display advertising[C]//2013 IEEE 29th International Conference on Data Engineering (ICDE). IEEE, 2013: 386-397.[2] Liu H, Pardoe D, Liu K, et al. Audience expansion for online social network advertising[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2016: 165-174.[3] deWet S, Ou J. Finding Users Who Act Alike: Transfer Learning for Expanding Advertiser Audiences[C]//Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2019: 2251-2259.[4] Zhuang C, Liu Z, Zhang Z, et al. Hubble: an Industrial System for Audience Expansion in Mobile Marketing[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2020: 2455-2463.[5] Liu Y, Ge K, Zhang X, et al. Real-time Attention Based Look-alike Model for Recommender System[C]//Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2019: 2765-2773.