A Study Of Graph-parallel Based Collaborative Filtering Location Recommendation Algorithm On LBSN

The recommender system can partially alleviate user's difficulty on information filtering and discover valuable information for the active user.As a result,recommender system has been considered as a typical method to deal with the over-loading of internet information.Collaborative filtering algorithm has the advantages of domain independent and supports users' potential interests.With the popularization of smartphone,location-based social network(LBSN)service become a trend.LBSN could combines social network information with the real world information.For these reasons,collaborative filtering has been applied to the location recommendation on LBSN.But,because of the dataset on LBSN has a multidimensional construct and in large-scale,recommender system on LBSN are facing big challenges of precision and performance.In order to solve the above problems,the main contribution of this dissertation are as follows:(1)This graduation thesis puts forward a GK-CF algorithm.By building a graph-based rating data model,the traditional collaborative filtering,graph algorithms and improved KNN algorithm have been integrated.Through the message propagation in the graph and the improved user similarity calculation model,candidate similar users will be selected firstly before the calculation of users similarity.Based on the topology of bipartite graph,the GK-CF algorithm ensures the quick and precise location of the candidate items through the shortest path algorithm.(2)And then,based on the effectiveness of the GK-CF algorithms,this thesis proposed a collaborative filtering algorithm LGP-CF on LBSN.The LGP-CF algorithm take the spatial-temporal information into consideration.In order to reduce the computation complexity,The user check-in dataset has been divided into three parts in accordance with the user historical Check-in pattern.Through cluster in the reconstructed data,candidate similar users will be selected firstly.The user similarity would be calculated by meshing the similar users' track data and Point Data.Then the candidate location would be selected repidly by location-based clustering.(3)Under the graph-parallel and data-parallel framework,GK-CF and LGP-CF have been parallelized design and implement.The extensibility and real-time performance of these algorithms has been improved,passed the performance tuning and process optimization of the algorithms.The experiments on real world clusters show that:compared with other collaborative filtering algorithms,the GK-CF and LGP-CF algorithms can better improve recommendation precision and the rating accuracy.The algorithms also has good scalability and real-time performance.