Abstract:
Nowadays, location-based social networks (LBSNs) have advanced to facilitate users in quickly sharing their check-ins
and rating points of interest (POIs), aiding in better targeting the preferences of future users. However, the variety of POIs and the
increasing interactions of LBSN users with these locations diminish the accuracy of memory-based recommendation methods (including
collaborative filtering, context integration, etc.), especially when dealing with new users. Furthermore, these methods also demonstrate
limited effectiveness, even when employing models such as matrix factorization, deep learning, etc., primarily due to the rapid evolution
of the history of LBSNs. To tackle these challenges, this article introduces a POI recommendation system (RS) that relies on implicit
trust among users within an LBSN. The system aims to (1) enhance the accuracy of recommendation methods through collaborative
filtering and (2) provide an alternative to explicit trust models that involve active user participation. This RS utilizes the HRCT (Hybrid
Rating Check-in Trust) model to deduce implicit trust from POI ratings and user check-ins, employing three types of trust matrices:
the TDMR (Trust Derivation Matrix based on Rating), the TDMC matrix (Trust Derivation Matrix based on Check-in) and the H-Trust
matrix combining these two matrices. The preliminary experimental results obtained with this model reveal that its algorithms achieve
better performance in terms of RMSE and Precision/Recall compared to collaborative filtering techniques using Pearson, Cosine and
Jaccard similarities. Moreover, this model can effectively address the data sparsity challenge of user/user similarity matrices by enhancing
the density of the model’s trust matrix derived from the H-Trust algorithm.
