Guiding Intelligent Ad Text Creation in Online Advertising

Subway hallway with advertising

Abstract

Online advertising is everywhere nowadays. People used to obtain information and find answers on search engines such as Google, Baidu, and Bing. Whenever a user puts a keyword into a search engine, relevant advertisements will appear on the search results page. This provides an opportunity for advertisers to present their products or services directly to potential customers. Also, social media platforms such as TikTok, Facebook, and Instagram play an important role in online advertising. These platforms can accurately place advertisements through user interest, attention and interaction data since they have huge user bases who browse and share content every day. This enables advertisers to present advertisements to specific target audiences, increasing the exposure and influence of advertisements.

But creating effective online ad text can be a daunting task for marketers and advertisers. With the advancement of artificial intelligence, we aim to use artificial intelligence to help marketers and advertisers to make their work more convenient.

In our thesis, we develop models for two use cases. The first model can output different styles of advertising text through the keywords input by the user, and the second model can output personalized marketing emails through the keywords input by the user. Users can get inspiration from these various advertisement texts and personalized marketing emails. The model we use for training is a pre-trained transformer model, Bloom-1b7 [DRR22]. Thanks to a large number of parameters of the pre-trained model, the model is still suitable for products in different fields even when the training data set is limited.

We conducted a user study and concluded that users were completely indistinguishable from ads generated by our model from those written by human marketers and that users preferred ads generated by our model. In terms of personalized emails, users prefer the personalized marketing emails generated by our model.


Junzhe Tai

Bachelor's Thesis

Status:

Completed

JavaScript has been disabled in your browser