基于改进Transformer的广告点击率预估模型

针对现有的广告点击率预估模型未能精准挖掘用户历史兴趣及历史兴趣对目标广告点击与否的影响,提出了一种基于改进Transformer的广告点击率预估模型.该模型采用Transformer网络捕捉隐藏在用户点击序列背后的潜在历史兴趣;同时针对Transformer建模用户历史兴趣无法有效关联目标广告的问题,提出了一种改进的Transformer网络.改进后的Transformer不但有效建模用户历史兴趣,而且考虑了跟目标广告的关联.新模型采用辅助损失函数来监督改进的Transformer对用户历史兴趣的抽取过程,然后采用注意力机制进一步建模用户的历史兴趣和目标广告的相关性以提升模型的预估性能.实验结果表明新模型有效提升了广告点击率的预估效果.

Improved Transformer based model for click-through rate prediction

Aiming at the problem that the existing click-through rate prediction models fail to accurately dig out the historical interest of users and the influence of historical interest on whether or not the target advertisement is clicked, this paper proposed a click-through rate prediction model based on the improved Transformer. The new model used Transformer network to capture the potential historical interest hidden behind the user''s click sequence. At the same time, in response to the problem that the Transformer modeling user''s historical interest sequence couldnot effectively associate the target advertisement, it proposed an improved Transformer network. The new model not only effectively captured the user''s historical interest, but also considered the association with the target advertisement to enhance estimated performance. Experimental results demonstrate that the new model shows better performance than other models.