| Abstract [eng] |
There is a huge amount of data generated on social networks on daily basis. Data from social networks and GPS data collected from users' mobile devices makes a Geo-Social Network. In this paper Geo-Social queries are implemented to analyze data from Geo-Social network. This data can be used for finding target audiences for advertisements or selecting rescue teams for disaster recovery. A model for executing Geo-Social queries and algorithms for query processing system is presented in this paper. Functions for evaluating geographical and social distances are defined. For evaluation of social distance between users, structural equivalence measure is used in this paper. With the help of Geo-Social queries, location prediction model of next user check-in is created. Quality threshold clustering algorithm is used to create clusters of check-ins. Check-ins are clustered together to find familiar regions for each user in Geo-Social network. Finding familiar regions for each user allows to classify check-ins into performed near familiar areas and performed in new ares. In this paper user's movement among clusters is modeled with Markov Chains model. An implementation of query processing system is presented by using real data from Gowalla social network. Location predictions are performed by different methods. First method uses user's location to predict location of next check-in. In another method location data of user's friends is used to make predictions. Third method combines user's location and his friends location data for prediction making. Finally, user's location and time of check-ins is used to make predictions. These methods are compared showing, that different method allows to achieve different prediction success rate. Finally, algorithms to make location recommendations for users in social networks are presented. Recommendations providing is tested on different types of users. Results described in this paper show, that Geo-Social queries are a useful tool for looking for insights in Geo-Social data. Implementation of Markov chains model in Geo-Social querying gives good results for predicting location of user's next check-in. Also Analysis of data from Gowalla social network show that users are making enough check-ins for using this data in Geo-Social queries. Results of simulations of recommendations making model show that accuracy of recommendations varies when different model parameters values are used. |
