User Location Preference Modeling via POItag Matrix Factorization

The patent introduces POItags as an intermediary element that mediates between users and locations. Instead of directly modeling complex user-location relationships, the system uses POItags as a bridge, creating user-POItag and location-POItag relations that can be factorized. This intermediary approach enables accurate preference determination through matrix factorization while maintaining manageable system complexity by breaking down the complex relationship into simpler components.

The patent applies local quality by allowing different locations within the same spatial cluster to have distinct POItag profiles. Instead of treating all locations in a cluster uniformly, the system enables each location to be characterized by its specific POItags, which capture local properties and user preferences. This allows reliable identification of individual locations within clusters by their unique tag compositions while still benefiting from the organizational efficiency of spatial clustering.

The patent transforms user and location data into a semantic space through probabilistic matrix factorization, changing the parameter representation from raw visit counts to latent factor vectors. This parameter transformation enables the system to capture semantic relationships and user preferences in a compressed form, improving interpretability and recommendation quality while the efficient factorization algorithms keep computation time manageable.