Points Of Interest Recommendation Algorithm Research Based On LBSN Implicit Check-in Data

With the popularity of smart mobile devices and the development of the Global Positioning System(Global Positioning System,GPS),people can access real-time location information more quickly and easily,which makes location-based social networks more and more popular.For example,Foursquare,Gowalla,Brightkite,etc.,these real-time online systems encourage users to share their life experiences in a circle of friends and check-in real time when visiting interesting locations.Our life movements are not only influenced by our colleagues and friends,but also refer to the points of friends on social networks,which will make location-based social a social trend that creates opportunities to motivate people to explore unknown places of interest.Although the research of location-based social network(Location-based Social Network,LBSN)recommendation system has been extensively studied,most of the research work and methods focus on the scene of displaying feedback data,but neglect the research based on implicit feedback data.However,in today's developed network,people can check-in anytime,anywhere.However,most of the check-in data generated is implicit feedback data,which only contains some basic check-in information(tour history,time,location)and check-in history.Most check-in data lacks explicit user preference information(such as preferring a place,not like a place),which in turn leads to complex recommendations based on implicit feedback.In this thesis,by mining the characteristics of implicit feedback check-in in location-based social networks,this thesis studies the recommendation of interest points based on implicit check-in.The main contributions of this thesis are as follows:1.This thesis introduces a geographic spatial-temporal distance measurement model,maps temporal and spatial information into a three-dimensional elliptical spherical coordinate system,maps temporal distance and spatial distance to a unified scale for measurement,and uses kernel function weighting to give different weights to spatial-temporal distance.Measuring the spatial-temporal distance under the same reference standard helps to alleviate the problems caused by cold start and data sparseness for position recommendation accuracy.At the same time,based on the Bayesian personalized ranking,a novel weighted Bayesian personalized ranking model based on spatial-temporal distance metric is designed,which has a good effect on the influence ofspatial-temporal transformation on the migration of interest points.Experimental verification on both datasets Brightkite and Gowalla performed well and was significantly better than the benchmark method.2.By analyzing the characteristics of implicit feedback,this thesis proposes a general location recommendation framework based on spatial-temporal context,personal preference and social relationship context under time effect to capture the characteristics of user's mobile behavior in implicit feedback.For the spatial distribution,the hierarchical clustering algorithm is used to perform distance-based hierarchical clustering on spatial locations,and different sub-classes are generated.Then,the kernel density estimation algorithm is used to calculate the kernel density estimation of the k-th position cluster.For time series analysis,the time is divided into fine-grained,according to people's living habits,the time is divided into the cycle mode in days and weeks,and then divided into different time periods,divided into working days and non-working days every week.For the influence of social friends,generate an adjacency matrix of friends related to the user,and use Jaccard to calculate the similarity of friends.Finally,a unified interest point recommendation framework is proposed,which combines spatial location distribution correlation and time-person-social friend correlation under time effect.