The LBSN POI Recommendation Algorithm Based On MapReduce

The rapid development and rapid spread of mobile Internet make people develop a habit of surfing the Internet anytime and anywhere, and the rapid development and application of 4G technology make this surfing habits to be consolidated and strengthened, which also gave birth to a new Internet business model: Location-based Social Networking Service. Location-based Social Networking Service providers provide check-in service to users on the social network platform, which is a location sharing service. After a social network user published his own location, his friends can see this location and the businesses represented by this location on the social network platform. Such behavior would have some impact on his friends’ consumer decision-making. For this reason, when a social network service provider has collected enough check-in data, it can use the personal preference and friendship to recommend location to users. Current location recommendation algorithm mainly use collaborative filtering method to recommend place of interest to users combine with user’s check-in location and check-in time. However, the density clustering method used for clustering check-in points often has a high time complexity, and the processing of time dimension is relatively coarse, so this article improves the traditional point of interest recommendation algorithm in the two aspects of running speed and recommendation accuracy.Firstly, the grid division thought is cited to improve DBCSAN density clustering algorithm, which makes the clustering object is no longer a single check-in point, but one grid that contains a lot of check-in points. Each grid will be judged whether it is dense grid or not after loading the points belong to it, which has greatly reduced the time to determine whether an object is a dense check-in point, and the number of clusters to traverse has been greatly reduced. In the handling of time dimension, this paper divides time dimension into several time slots according to people’s daily routine, instead of the traditional continuous time clustering. This improvement makes the calculation of user similarity more accurate. Finally, this paper uses the MapReduce parallel programming model to improve the clustering algorithm based on grid, so as to further improve the speed of recommended algorithm. The experimental comparision of traditional point of interest recommendation algorithm, Grid-based point of interest recommendation algorithm and recommendation algorithm based on MapReduce proves that this paper’s improvements can improve the speed and accuracy of recommendation.