The emergence of Location-Based Social Networks has made significant revolution in spatial information technology. These Networks are one of the best enriched sources for supplying various location based services by mobile devices due to spatial data acquisition and updating from all around the globe. One of the most important and most applied location-based services are way finding service. Way finding is one of the common problems in present life for conducting Mobile Agent to its destination without feeling lost. Using landmarks is the most effective approach for way finding in an unfamiliar environment. In this research, landmark dataset and local users are extracted from geotagged photos with text and data mining with creating Location-Based Social Networks. In addition, the graph is designed for optimal route determination that its nodes are landmarks and crowded road segmentation which, frequently traveled by local users, and its edges are determined by ordering those road segmentations. According to this graph, two steps of routing are used to obtain the optimal path. To implementation, the used data are gathered from 22 vehicles trajectory in Tehran city for 3 months. Finally, a recommendation system has been implemented for recommending place of interest as personal landmarks to facilitate user orientation in searching region. To evaluate performance of the two-step routing, we experimentally compared the waste of time in proposed routing with Dijkstera shortest path for different lengths and users with different levels of regional knowledge.