Research On Individual And Group POI Recommendation Based On Social Networks

With the rapid development and popularity of GPS and the mobile Internet,social networks face a great change.Location-based social networks(LBSN)and event-based social networks(EBSN)as two representative types of social networks have become more and more popular.More and more people are willing to share their check-ins and participate in various activities,which leads to a huge number of social data with geographical information.In order to effectively mine valuable information from these data and serve to users,recommendation technologies play an import role and have attracted extensive attention from businesses and academia.This paper mainly focuses on POI recommendation,in order to respectively recommend potential POIs to individual users and groups.By more deeply analyzing the nature of problems and utilizing multi-dimensional information in social networks,this paper proposes several more reasonable and effective recommendation methods to provide better recommendation service.The main research works in this paper are as follows:(1)Analyze background significance and related research status at home and abroad of POI recommendation,and then discuss the main ideas of traditional recommendation algorithms with advantages and shortcomings to provide a theoretical basis for the next research work.(2)Discover and analyze the hierarchical structures in users' check-in behaviors.On account of this phenomenon,based on the existing GeoMF model,this paper proposes a hierarchical geographical matrix factorization method,and explores the content and spatial implicit hierarchical structures of both users' preferences and POIs to improve the performance of individual POI recommendation.(3)To solve the problem that how to make a well trade-off among different members' preferences in the target group,this paper proposes a novel group recommendation generative model.This model considers the fact that personal preferences,group topics and their social relations have an influence on their action.When making POI recommendation for the target group,add each member's topic-dependent influence weight to target group,and aggregate different preferences of all members in group to get the whole preference of a group to a candidate POI.Finally,this paper respectively conducts comparison experiments on real-world datasets for individual and group POI recommendation.Compared with existing methods,two methods proposed in this paper both show more superior performance in terms of several evaluation metrics,which proves the efficiency and reliability of the proposed methods.