amadeus-research-publications

Deep Choice Model Using Pointer Networks for Airline Itinerary Prediction

← Back to Publications

Publication Details

• Authors: Mottini, A.; Acuna-Agost, R.
• Venue: ACM SIGKDD 2017
• Year: 2017
• Link: ACM DL

Abstract

This paper presents a novel deep learning choice model for predicting which airline itinerary a customer will select, using a sequence-to-sequence architecture called Pointer Networks. Unlike traditional discrete choice models, the pointer network can directly handle variable-length sets of flight options and model complex interactions (via its attention mechanism). The model was trained on large-scale airline booking data and achieved higher accuracy in predicting passengers’ choices compared to baseline models. This demonstrates the potential of deep neural network architectures in choice modeling, especially for applications like itinerary ranking and personalized offer optimization in the travel industry.

Key Contributions

• Application of pointer networks to choice modeling
• Deep learning approach for airline itinerary prediction
• Novel architecture for handling variable-sized choice sets
• Strong empirical results on real-world travel data

Citation

@inproceedings{Mottini2017DeepChoice,
  author    = {Andrea Mottini and Rodrigo Acuna-Agost},
  title     = {Deep Choice Model Using Pointer Networks for Airline Itinerary Prediction},
  booktitle = {Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD 2017)},
  year      = {2017},
  publisher = {ACM}
}

---

← Back to Publications