The optimization of keyword generation, expansion and selection in housing brand advertisement creativity has a role to play in the enhancement of brand awareness in the housing market that should not be taken lightly. In this paper, a generative pretraining model (ProphetNet) containing a future prediction mechanism is used as a prediction method for advertising keywords, which effectively enhances the model’s understanding of the context and the coherence of the generated text by predicting multiple future words. After obtaining multiple ad creative keywords, the hierarchical Bayesian model is used to summarize the prior information of ad keywords based on historical data and experience. And the parameter estimation between the ad creative keywords and the brand recognition in the housing market is performed to generate the final keywords. The designed ProphetNet model consistently stabilizes the average accuracy at 0.6 and above under a variety of keyword expansion numbers, which is both effective and stable.
