Comparison of matrix factorization algorithms' performance in tourist attraction recommendations.

In recent years, the vast amount of offered items can overwhelm users in processing the information for decision-making purposes. In this case, recommender systems have become a functional tool to help people make good decisions. Big companies such as Netflix, Spotify, and YouTube have employed recommender systems in their platform. In the tourism industry, the case also exists, especially when travelers visit a tourist city such as Paris or Bali, which offers them many tourist attractions. In this paper, we employ collaborative filtering techniques to model a recommender system for tourist attractions. Singular value decomposition (SVD), SVD++, and non-negative matrix factorization (NMF) algorithms are used in this paper over a dataset which we collected previously from 44 users and 60 tourist attractions. The optimal parameters of the algorithm are searched based on exhaustive searching techniques against a set of parameter combinations. We used 5-fold cross-validation to evaluate the predicted rating in each algorithm in terms of Root Mean Square Error (RMSE) and Mean Average Error (MAE). Results show that using our dataset, the model built in SVD++ had the minimum error but took a longer training time, while NMF gave the worst RMSE but had a faster training time compared to SVD++. [ABSTRACT FROM AUTHOR]