Semantic Tagging of Places Based on User Interest Profiles from Online Social Networks

Abstract

In the recent years, location based services (LBS) on mobile devices have become very popular. With the growing number of smartphone users, the demand for services that can provide recommendation of places based on user location and interest has increased rapidly. The performance of such LBS depends on a number of factors, including how well the places are described. A number of location based services allow users to check-in at places, i.e. users can let others know of their whereabouts. Even though they also enable users to manually tag places they have visited, users rarely do so. Moreover, the available information attached to places (e.g. their names) is often ambiguous or insufficient for service providers to automatically generate tags. On the other hand, users often provide information about their interests in online profiles via online social networks. The common interests of a group of people that has visited a particular place can potentially provide further description for the place. In this work we present an approach that automatically assigns semantic tags to places, based on interest profiles and check-in activities of users. The approach consists of: (i) an interest profile expansion algorithm to derive semantic concepts related to the user interests; (ii) a model to determine the probability that a particular semantic concept describes a place, based on the check-in activities of users; and (iii) a noise removal approach, using a hierarchical clustering technique, which is applied on the top probable semantic concepts to derive the final semantic tags for places.We have evaluated our approach with real world datasets from popular social network services, against a set of manually assigned tags. The experimental results show that not only we are able to automatically derive meaningful tags for different places, but also that the sets of tags assigned to places are expected to stabilise as more unique users check-in at places. This indicates that top probable tags derived can be consistently assigned to places irrespective of the number of people who have checked-in at those places.