LOCATION BASED RECOMMENDOR SYSTEM USING ENHANCED RANDOM WALK MODEL

Abstract

The increasing adoption of location-enabled smartphones has given rise to a number of social services (such as Foursquare and Google Places) that people can use to trace, annotate, and share their experiences about the locations they visit as they navigate through their daily lives. Users notify their friends of the place where they are with a check-in, leaving a digital trace of their movements. These services therefore now hold and collect huge datasets that can location users’ mobility to both their social connections and the spatial layout of their cities; providers are also beginning to apply a host of machine learning approaches to their data in order to turn check-in services into fully fledged location recommendation systems. First, location recommendation can help users to find potential friends, a function that improves user experience in social networking and attracts more users consequently. Compared with the usual passive ways of locating possible friends, the users on these social networks are provided with a list of potential friends, with a simple confirmation click. Second, location recommendation helps the social networking sites grow fast. A more complete social graph not only improves user involvement, but also provides the monetary benefits associated with a wide user base such as a large publisher network for advertisements. This work proposed a framework using both attribute and structural properties to recommend potential location in social networks. To compute accurate location recommendations in social networks, we propose a list of desired criteria. A random walk framework on the augmented social graphs using both attribute and structural properties is further proposed, which satisfies all the criteria. We also discuss different methods for setting edge weights in the augmented social graph which considers both global and local characteristics of the attributes.