Research On Link Prediction Based On Transactional Information

Link prediction is a new research subject full of challenge. Along with the rapid development of Internet and the spread of electronic application, which produced more and more large scale online social network data based on Internet, these data collect easily and reflect the connected relationship among people on the net from different ways. The main study object of the paper is based on the specific online social network. According to the characteristics of the network and the main activities among people, we view the interactions between entities as transactional activities in the network, and improve the existing link prediction algorithms to perform link intensity prediction based on the transaction information between entities. This paper studies the contents can be divided into the following several aspects. In allusions to this problem, this paper does some research as followed.Firstly, because of the repeat occurrence between entities in the online social network, this paper construct different network graphs firstly according to the main function of the network, considering the transactional information between entities and the attribute information to extract suitable features, and then adopt the graph similarity methods and the supervised learning algorithm to perform link prediction, to improve the accuracy of the prediction links.Secondly, one significant feature of the online social network is the network pattern would change with time. In order to perform link prediction, in this paper, based on analyses the structure of online social network, we bring forward a link prediction framework based on the social network, this paper extract temporal features of the network first using the decay kernel functions, and then use the extended relational bayesian classifier to perform link prediction, to improve the performance of the link prediction.Finally, this paper conduct the experiment validation and results for the proposed algorithms respectively. The experiment results show that the proposed algorithms have a high accuracy, and achieve expectant research aim.