Check-in User Operation Scene Recognition Based on Deep Temporal Perception Network

With the acceleration of digital transformation in the air transport industry, the traditional manual check-in mode is facing dual challenges of efficiency bottlenecks and service experience upgrades. However, the time dependence and diversity of check-in user behavior make it difficult for traditional classification models to cope with their complex non-linear and temporal characteristics. To this end, a deep learning model for check-in user behavior classification that combines the bidirectional gated recurrent unit (BiGRU) with the Transformer is proposed, named BiGRU-Transformer. The proposed model can capture the bidirectional dependency relationships in the time series through BiGRU, and utilizes the multi-head attention mechanism of the Transformer to enhance the ability to identify and focus on key features. It not only improves the classification accuracy of the model for future check-in user behavior, but also shows significant advantages in handling the complexity and variability of data. The experimental results based on the travelsky_with_type dataset show that this model has excellent performance in classifying check-in user operation scene.