Abstract:
With an overwhelming number of travel options available online, users often struggle to
identify destinations that align with their preferences, interests, and constraints. Traditional
travel platforms lack personalization, often providing generic suggestions that do not consider individual travel histories or specific criteria such as budget, preferred climate, or desired activities. This results in decision fatigue and reduced user satisfaction. The project aims to address these challenges by developing a personalized travel destination recommendation system that delivers relevant and tailored suggestions.
To tackle the problem, a hybrid recommendation approach was implemented, combining
collaborative filtering and content-based filtering techniques. Collaborative filtering was
implemented using SVD, which decomposes the user-item interaction matrix to uncover latent features that represent user preferences and destination characteristics, enabling effective recommendations even with sparse data. Content-based filtering utilized NLP techniques, specifically BERT embeddings, to analyze and compare destination features such as type, climate, budget, and attractions with user preferences.
Functional and non-functional aspects were clearly defined and evaluated throughout
development. The novel hybrid model achieved Accuracy of 0.80, Precision of 0.77, Recall of
0.79 and F1-Score of 0.78. The results demonstrate the potential of the hybrid recommendation system in accurately capturing user preferences and delivering relevant travel suggestions, even across diverse user profiles and sparse datasets. By leveraging the strengths of both content based and collaborative filtering techniques, the system maintains recommendation quality while effectively addressing challenges such as cold start and data sparsity.
