Transformer-based Multi-Touch Attribution Model for Large-Scale Digital Advertising Paths

As digital advertising becomes more complex and widespread, traditional multi-touch attribution models and statistical methods often struggle to achieve high accuracy. While time-series deep learning models offer improvements, they face challenges such as long training times, slow parameter updates, and limited parallel processing. In response, this paper introduces the Transformer-based Multi-Touch Attribution model (TRMTA), which uses the Transformer’s global attention mechanism to enhance attribution accuracy in large-scale digital advertising campaigns. We validate the TRMTA model using nearly 260 million touchpoint data points from splash ads across 10 platforms. The results show that TRMTA outperforms traditional models like LSTM and Logistic Regression (LR). In real-world applications, the TRMTA model has helped advertisers optimize their strategies for the Double Eleven (11.11) shopping festival, resulting in a revenue increase of over $10 \%$.