Click-through Rate Prediction Model Based on Dynamic Graph Attention Mechanism Network
Houchuan Yuan, Chengwan He · 2021
Most of the current models are based on the user’s specific historical behavior for interest mining, when the user behavior is sparse, it is difficult to capture the user’s intentions, resulting in insufficient predictive power and other problems. Therefore, a prediction model based on Dynamic Graph Attention Network (DGAT) is proposed. This model uses a recurrent neural network to model dynamic user behavior, and uses a graph neural network to model the user’s social impact, which enriches user behavior and solves the problem of behavior sparseness. In addition, the local activation unit and adaptive attention mechanism are introduced into the model to enhance the generalization ability of the model. Experiments have proved that the DGAT model outperforms the existing models in various indicators.