Abstract:
In recent years, tourists have tended to share their travel experiences with friends through location-based social networks
(LBSNs). However, these networks accumulate large masses of data, making them ineffective in guiding individual tourists through their
journeys. To overcome this drawback, point-of-interest (POI) recommender systems (RS) can provide a beneficial solution by exploiting
the potential of LBSNs to suggest places they have never visited to new tourists. These systems can be classified into two categories:
the first uses memory-based algorithms, while the second employs algorithms based on machine learning models. Collaborative filtering
(CF) is a popular memory-based smart tourism approach commonly used in literature. This approach predicts the probability of POI
check-ins by new tourists based on their similarities with other tourists, using measures such as Cosine, Jaccard, Pearson correlation,
and Euclidean distance. However, to our knowledge, no formal framework takes POI check-ins and visit paths into account when
calculating similarities between tourists. For this reason, in this paper, we propose a novel measure called SPPUR (Similarity of Paths
and the Proximity of Users for Recommending POIs) inspired by the term frequency-inverse document Frequency (TF-IDF) method,
which uses POI frequentation and geographical proximity between users to calculate similarities that can predict POIs to be visited by
new tourists. Our experimental results on Foursquare show that compared with other state-of-the-art measures, this similarity measure
significantly improves SR performance regarding PRECISION, RECALL, MAP, and NDCG
