Towards a systematic point-of-interest recommendations based on trust between users deduced from their ratings and check-ins in a LBSN

Nowadays, Location-Based Social Networks (LBSNs) enable users to swiftly share their evaluations of Points Of Interest (POIs), helping to better predict and cater to the preferences of future users. The wide variety of POIs, biased ratings, and the challenges of training POI recommendation models (e.g., matrix factorization, deep learning) in the dynamic context of LBSNs reduce the accuracy  of POI rating predictions, particularly when addressing the needs of new users. To address this issue, memory-based recommendation methods that leverage explicit trust declared by users can offer a viable solution, as they often provide greater accuracy than approaches based on user-user similarities. However, these methods are hindered by the cold start problem, as most users do not specify their trust relationships. For these reasons, this paper proposes a model named HRCT (Hybrid Rating Check-in Trust) to infer implicit trust from POI ratings and user check-ins. This recommender system (RS) aims to (1) alleviate the cold start problem in the context of LBSNs and (2) provide an alternative to  explicit trust models that require active user participation. To achieve these goals, the HRCT model uses three types of trust matrices: the Trust Derivation Matrix based on Rating (TDMR), the Trust Derivation Matrix based on Check-int(TDMC), and the H-Trust matrix combining these two matrices. Preliminary experimental results obtained with this model reveal that its algorithms achieve acceptable performances in terms of RMSE and Precision/Recall compared to collaborative filtering techniques using Pearson, Cosine, and Jaccard similarities. Additionally, this model effectively tackles the sparsity challenge by increasing the density of trust matrices by 2.95% compared to data from user-user similarity matrices.