The Research And Application For Link Prediction In Complex Networks

In this thesis,the main research contents are complex network characteristics and link prediction algorithms,the main purse of the study is to improve the predictive effect.And it is expected to apply the structure characteristics of complex networks to the link prediction to improve the predictive accuracy.Firstly,the experiment is carried out to simulate the construction of WS small world model and BA scale-free model.Based on the experimental results to analyze the relationship between clustering coefficient and reconnection probability or network size.And it is found that the formula of clustering coefficient in WS small world model depends on the premise of network size.Secondly,make the index of power for the existing similarity index algorithm which based degree of nodes.By calculating different prediction effect under the different exponent,it is found that the effect of the intermediate node's degree on the prediction effect is much higher than that of the terminal node's degree.Also by improving the RA index algorithm and further differentiate the different neighbor node's contribution to the two end nodes a new CRA index algorithm is put forward.At the same time by increasing the influence of end nodes degree,CRA-? index algorithm is proposed.Through the repeated experiments we can find that compared with the other similarity index algorithm,the prediction effect of CRA index in multiple networks have been got different level of ascension.Also by choosing the appropriate ? value,making the prediction effect of CRA-?index compared with CRA index further ascension.Thirdly,Moreover through comparing the network structures of multiple data sets,select the several common basic network models and analyze the prediction effects of different similarity index algorithms in different network models.According to the community structure,put forward a relatively novel LSCN index algorithm based on the structure similarity of neighbor nodes.Through the neighbor node's structure similarity to calculate the connection probability,and then predict the probability of connection between two nodes.And by the final experiment results can be seen,compared with other algorithm based on node structure similarity the predict effect has been improved significantly.After LSCN-? index algorithm is proposed,by further considering the second order structure similarity of neighbor nodes.And choose the optimal parameter makes the final prediction effect get a further ascension.Finally,use edge betweenness of complex network to replace the edge weights.To a certainextent,solved the problem of weight loss in the weighted network and the problem of the weight calculation method is not unified.And through the experiment further validated the effect of weak connection in weighted networks.Also,by the final experiment results can be seen,use edge betweenness to replace the edge weights can improve the predictive effect in weighted networks.