Institutional Repository [SANDBOX]
Technical University of Crete
EN  |  EL

Search

Browse

My Space

A novel hybrid recommender system for tourism

Ziogas Ioannis-Panagiotis

Full record


URI: http://purl.tuc.gr/dl/dias/11779481-392F-4BF8-BF5D-979873946FF7
Year 2023
Type of Item Master Thesis
License
Details
Bibliographic Citation Ioannis-Panagiotis Ziogas, "A novel hybrid recommender system for tourism", Master Thesis, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2023 https://doi.org/10.26233/heallink.tuc.96217
Appears in Collections

Summary

In this MSc thesis, we put forward several novel recommender algorithms integrated into a hybrid recommender system for the tourism domain. To this end, we first explore the use of semantic similarity measures for Content-based recommendations to suggest tourist attractions. We study ways of deploying hierarchies of points of interests (POIs) and operate upon them with well-known similarity measures originating in the text analysis domain. Then, we progressively build three novel, hierarchy-free, similarity measures and discuss their strengths and weaknesses. We end up with a measure, the Weighted Extended Jaccard Similarity (WEJS) that combines information regarding the user interests (in the form of user preference-related weights) and specific item’s characteristics (in the form of particular values for the item’s features). As such, the use of WEJS allows the provision of recommendations that are effectively personalized. Interestingly, though it is a hierarchy-free measure, it is able to recommend items based on others that would naturally appear close in a feature-based POIs hierarchy; while at the same time it is capable of capturing similarities among items that would be distant to each other in any hierarchy build solely based on the POIs’ features. Our systematic experimental evaluation using real-world data showcases the benefits and limitations of the various measures and confirms the effectiveness of WEJS in offering “rich” and personalized recommendations, so that it can be utilized as important sub-component of a complete Recommender System. Subsequently, we develop two novel recommender algorithms, a Content-based one and a Hybrid which combines (a) a Bayesian component used for eliciting user preferences, and (b) the aforementioned Content-based algorithm as recommendations component. The second component can in fact itself be considered a hybrid among two different algorithms exploiting semantic similarity measures: a hierarchy-based and WEJS, the non-hierarchy based one. We evaluate our approach via extensive simulations conducted on a real-world dataset constructed for the needs of a real mobile application for short-term visitors of the popular touristic destination of Agios Nikolaos, Crete, Greece. Our experiments verify that our algorithms result in effective personalized recommendations of touristic points of interests; while our final hybrid algorithm outperforms our exclusively Content-based recommender algorithms in terms of recommendations accuracy. Parts of the research results produced in this thesis appear in three scientific articles, two published after peer-review in a scientific journal and the proceedings of an international conference, and one currently under review.

Available Files

Services

Statistics