Link Prediction In Social Network Based On Probabilistic Model

It is helpful for us to study link prediction in social networks because of its contribution to understand potential relationship between two nodes more deep and to analysis and predict nodes. In view of this, based on the complex network theory knowledge, we build the corresponding improved Hawkes model and supervised random walk model according to the structural characteristics of social network thus making its simulation to replace real social network nodes and providing a powerful research platform for prediction further. The main contents of this thesis are as follows:(1) Aiming to the present forecast model less consider influence factors and time attributes. In this thesis, the Hawkes model is introduced to predict the possible nodes of the network. In our topic, first of all the calculation method of node influence the relationship between nodes are analyzed, then we can find the key nodes. Secondly we introduce intensity function in the Hawkes model, then calculate strength between the new node and other nodes. Finally, we calculate probabilities between nodes and rank them from large to small, the greater node have greater probability. The experimental results show that the algorithm in speed and accuracy rate are significantly much better than Hawkes model without influence function.(2) The user behavior is also an important factor in social networks. Because Hawkes model method fused with the influence function less considered user behavior. Supervised random walk algorithm can make full use of the node's attributes and behavior information. So we have improved the random walk algorithm and added the supervised method to solve the problem that the user behavior factors are less. Firstly, the characteristics of the nodes are classified, such as gender and preferences and so on. Secondly the parameters are used as the parameters to train the training set. Finally, the optimal parameters are obtained to guide the random walk of the test set. This algorithm not only can further speed up the prediction of network chain edge, but also consider the adaptation of large scale network prediction. In the experiment, we have used improved Hawkes model method fused with the influence function and supervised random walk model and other classical method to predict the same real data, and then compared the results of the models. Results show that when the number of nodes more than 2500, the supervised random walk algorithm have better effect and simulate results verify its validity.