Research And Design On Adaptive Recommendation System For Social Users

Social network is a user centered and relationship connected platform,which maps the real world interpersonal relationship into a virtual network.With the help of Internet,social network connects people from different locations and different races,and provides a united platform for information sharing,interaction and other functions.Recommendation system has alleviated the problem of slow and low efficiency of getting information caused by the huge amount of data,the diversity of its structure and the complex relations in big data era.As one of the services provided by recommendation system,friend recommending is significantly helpful in maintaining relationship and broadening contacts for users.Because of personality differences,a user may play different roles in social network,and various users need different friend recommending methods.Therefore,according to the analysis of the related research on social network and recommendation system at home and abroad in recent years,the thesis focuses on two aspects: hierarchical recommendation and adaptive recommendation system for social users.The main work and innovations are as follows:1.This thesis designs a hierarchical recommendation model based on bipartite graph.Firstly,this model simplifies the native social network structure by dividing overlapping communities and further constructs the bipartite graph consisting of communities and users.Secondly,based on this graph the model achieves role dividing by leveraging the topological features of bipartite graph and the attributes of nodes.Finally,based on bipartite graph consisting of two users,this model generates the friend recommendation lists,which results in hierarchical and personalized recommendation.2.This thesis studies and designs a social users oriented adaptive recommendation system.Different from the traditional way of using the same recommendation method for all social users and considering that friend recommendation methods applying for users with diverse category are not same,several recommendation algorithms are integrated into the recommendation system with a unified standard interface.The classifier for recommendation algorithm is generated based on users' profile through AdaBoost.When social users do behaviors,the recommendation system perceives the change and triggers the classifier to adaptively match or update users' current recommendation algorithm.Finally,the thesis verifies the designed model and system with douban data.