Recommender systems analyze conditions and user behaviors to recommend proportional services to users. Since the aim of such systems is to provide the most appropriate services, it appears essential to use filtering techniques to limit recommender items. In this study, spatial criteria such as distance, movement direction, visibility, and topological relationships were employed as filtering tools to provide the right items. Our model creates appropriate items for better recommendation based on spatial relationships between users and the surrounding service sites. This method demonstrates that the number of recommended items can be limited by considering the shortest distance from the service centers intended by users and taking user direction into account. Moreover, appropriate service centers can be proposed with respect to user visibility. In this study, topological relationships between user location and near places were used as spatial filters, too. Further, if these filters can interact with the environment in the same way as humans, it can be expected the recommendation process to improve. Thus, our model uses the fuzzy approach to help the system to perceive the uncertainty of the spatial linguistic terms. To evaluate the performance and effectiveness of our proposed spatial filtering model, we conduct several experiments on real datasets that were obtained from tracking the users’ location through GPS. Considering the actual conditions, this system solved the cold start problem using spatial filtering model. Experimental results show that 68% of test users considered our recommendations as relevant in new item cold start problem. Moreover, results reveal that compared with an LA-LDA model, using spatial filtering in cold start item problem is more robust.
Critique on Natural Noise in Recommender Systems
