Generating Search Engine Text Advertisements considering keyword insertion with BERT2BERT

Abstract

In sponsored search advertising, it is known that inserting search keywords into the ad text improves advertising metrics and Key Performance Indicators (KPIs), such as, in particular, the click-through rate. This paper proposes generating better ad text that inserts the search keywords with a high probability. We used BERT2BERT as the base model and implemented the keyword MASK model and keyword priority insertion algorithm. The keyword MASK model is a method that masks keywords while training to prevent the generation of similar words, and the keyword priority insertion algorithm is a technique that improves the probability that keywords appear in an ad text. Experimental results with actual delivered ad data show that the proposed method's insertion ratio achieves up to 80%, significantly higher than the baseline model's.