Research On Influence-Based Recommendation Methods On Geo-Social Networks

With the rapid development of mobile positioning technology and social networks,a large number of geographic social networks have emerged in recent years(including:location-based social networks,event-based social networks),such as Foursquare,Facebook Places,Gowalla and Plancast,Yelp,Meetup,Douban activities,etc.In geographic social networks,users can obtain current location information through smart terminals anytime and anywhere,and log in points of interest.Leave a personal footprint in social networks.The sharing function allows users to comment on current points of interest for sharing information and personal experience with friends.Geographical social network is a heterogeneous information network,including users and locations,as well as their mutual information.Therefore,this paper refers to it as Heterogeneous Geographical Social Network(HGeo SN).Heter Geo SN is a directed weighted heterogeneous information network with multiple types of nodes and link relationships.This new type of social network that covers many types of information,complex network structure,and diversified user activities in the network,the types and requirements of geographic social application services in the new social network have become more and more diverse and complex.Therefore,how to propose a recommendation technology suitable for this heterogeneous geographic social network that integrates heterogeneous information networks and diversified user activities is a problem that needs to be solved urgently in recent years.Starting from the “multi-type,semi-structured,non-display,and hierarchical” characteristics of heterogeneous geographic social networks,this paper conducts a systematic and in-depth study on the key technologies of recommendation in heterogeneous geographic social networks.The main research work are as follows:1.Research on the influence model of recommendation issues.Influence is a complex and subtle force that controls social dynamics and user activities in heterogeneous geo-social networks.Understanding how users and locations influence each other in heterogeneous geo-social networks can benefit a variety of applications,such as viral marketing,recommendation,and information retrieval.Previous research mainly used the local centrality of the network to quantify the influence,ignoring that the geographic social network is a information network,and the information in the network is transmitted in one direction along with the interaction between the user and the location.Therefore,thispaper mainly uses the topological centrality of heterogeneous geographic social networks to quantitatively learn the influence between users and locations in the network; fully considers the compound correlation,spreading and asymmetry of influence and user activities and behaviours in HGeo SN.This paper proposes the homogeneous influence models and the heterogeneous influence models.The influence model proposed in this paper is mainly based on the check-in behaviour of users in geographic social networks.These models are a quantitative analysis of internal and external factors that affect user activities.It has important theoretical significance for the study of recommendation problems based on user activity behaviour.2.Research on the method of the friend recommendation based on the influence.People rely on social media to meet two common needs(i.e.,social needs and information needs): to keep in touch with friends in the real world and to obtain information that interests them.The traditional methods of recommending friends on social media mainly focus on the social needs of users,and rarely pay attention to the information needs of users,so that the friends obtained by the target user through the recommendation system are invalid recommendations in terms of information acquisition.This paper proposes a friend recommendation algorithm that takes into account the needs of broadening the circle of friends and obtaining information.This paper first establishes an influence-based treestructured preference model for the behavioural preferences of users in the network; then,proposes a preference similarity metric model to measure the similarity between users;secondly,it proves our proposed friend recommendation problem is an NP-Hard problem.A friend recommendation optimization algorithm with preference coverage is proposed to reduce the complexity of the problem; finally,through a multi-objective optimization learning framework,the framework solves the sensitive issues of aggregation parameter.Experiments on real datasets show that the algorithm proposed in this paper is superior to the existing advanced friend recommendation algorithms in terms of information richness while ensuring good recommendation accuracy.3.Research on the method of location recommendation based on the influence.The recommendation system is composed of users,content providers(business),and platforms.However,existing research methods mainly focus on satisfying users’ service needs.Therefore,how to recommend a location for the target user while at the same time,the target user acts as a seed node for information promotion to promote the influence range of the recommended location(business)in the geographic social network.This is the key to solving the problem of the locality of existing research service objects.This paper takes into account the service needs of multiple participants and proposes a location recommendation algorithm driven by the business information promotion.This paper first proposes the heterogeneous activity schema of users; then,according to the complexity of the location recommendation problem,proposes a location-related preference model that is suitable for the service needs of multiple participants; secondly,the location recommendation is essentially to establish a link between the service object and the recommended location,and heterogeneous influence is introduced into the recommendation objective function to characterize the gain effect of increasing the links influence on the recommended location; finally,according to the type of service objects,we propose a user-oriented location recommendation algorithm and a businesses-oriented location recommendation algorithm.Experiments show that the algorithms proposed in this paper are superior to the existing advanced location recommendation algorithms in HGeo SN in terms of evaluation indicators such as accuracy and the gain of influence range.4.Research on the method of multi-type content recommendation based on the influence.Because people’s activity behaviour in geographic social networks is a directed sequence composed of time,activity type,and activity location,and as social animals,participating in group activities is an indispensable part of people’s daily social life.Therefore,it is an important task of the current recommendation system to recommend the activities represented by the directed sequence to the user and the partners who participate in the activities together.However,the existing recommendation services mainly focus on the recommendation of a single element in the activity composition.Therefore,this paper proposes a new recommendation paradigm: multi-type content recommendation for group activities,recommending a list of heterogeneous activity patterns composed of activities,activity locations,and activity schemas for service objects.First,the heterogeneous activity schema of user groups is proposed; then,in order to dig out the decision-making process of group activities,the group influence models and group preference models that affect group activity behaviours are proposed; finally,according to the different ways of forming user groups,this paper proposes multi-type content recommendation algorithm for users and user groups.The experimental results show that although the existing recommendation algorithms for group activities are mostly single-type content recommendation algorithms,comparing the recommended content of the algorithm proposed in this paper with the single-type content recommendation algorithm for group activities,the accuracy and the recommendation quality has improved significantly.