Recommending Tourist Attractions Using Neural Collaborative Filtering and Transformer-Based Models
(2025) DABN01 20251Department of Economics
Department of Statistics
- Abstract
- This thesis aims to create a personalised recommendation system for tourist attractions based on users’ individual preferences. Using a TripAdvisor dataset filtered down to ten European capital cities, two deep learning models are compared: neural collabora- tive filtering and its extended transformer-based architecture. To address the imbalance encountered in the dataset (where ratings of four and five stars are overrepresented), two additional datasets are generated using data augmentation and inverse probability sampling weighting, producing six tuned models. Model evaluation is performed using three evaluation metrics, mean absolute error, root mean squared error, and normalised discounted cumulative gain. The results demonstrate that... (More)
- This thesis aims to create a personalised recommendation system for tourist attractions based on users’ individual preferences. Using a TripAdvisor dataset filtered down to ten European capital cities, two deep learning models are compared: neural collabora- tive filtering and its extended transformer-based architecture. To address the imbalance encountered in the dataset (where ratings of four and five stars are overrepresented), two additional datasets are generated using data augmentation and inverse probability sampling weighting, producing six tuned models. Model evaluation is performed using three evaluation metrics, mean absolute error, root mean squared error, and normalised discounted cumulative gain. The results demonstrate that the Transformer-based model trained on the original dataset performs the best across all three evaluation metrics. This aids to highlight the model’s ability to capture patterns in the data and subsequently model the relationships between users and attractions successfully without the need for additional techniques to create a more balanced dataset. Furthermore, the function that generates the recommendations differentiates between known and cold-start users and al- lows for both to receive personalised recommendations. This underlines the potential of Transformer-based architectures to serve as a reliable alternative to neural collaborative filtering in the context of recommendation systems. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9193536
- author
- Totaro, Paolo LU and Akdemir, Dilay LU
- supervisor
- organization
- course
- DABN01 20251
- year
- 2025
- type
- H1 - Master's Degree (One Year)
- subject
- keywords
- Recommendation system, Neural Collaborative Filtering, Transformer, Deep Learning, Tourism
- language
- English
- id
- 9193536
- date added to LUP
- 2025-09-12 09:05:19
- date last changed
- 2025-09-12 09:05:19
@misc{9193536, abstract = {{This thesis aims to create a personalised recommendation system for tourist attractions based on users’ individual preferences. Using a TripAdvisor dataset filtered down to ten European capital cities, two deep learning models are compared: neural collabora- tive filtering and its extended transformer-based architecture. To address the imbalance encountered in the dataset (where ratings of four and five stars are overrepresented), two additional datasets are generated using data augmentation and inverse probability sampling weighting, producing six tuned models. Model evaluation is performed using three evaluation metrics, mean absolute error, root mean squared error, and normalised discounted cumulative gain. The results demonstrate that the Transformer-based model trained on the original dataset performs the best across all three evaluation metrics. This aids to highlight the model’s ability to capture patterns in the data and subsequently model the relationships between users and attractions successfully without the need for additional techniques to create a more balanced dataset. Furthermore, the function that generates the recommendations differentiates between known and cold-start users and al- lows for both to receive personalised recommendations. This underlines the potential of Transformer-based architectures to serve as a reliable alternative to neural collaborative filtering in the context of recommendation systems.}}, author = {{Totaro, Paolo and Akdemir, Dilay}}, language = {{eng}}, note = {{Student Paper}}, title = {{Recommending Tourist Attractions Using Neural Collaborative Filtering and Transformer-Based Models}}, year = {{2025}}, }