Research On Link Prediction Methods Based On The Similarity

Link prediction in social networks as an important branch of data mining is a sub-task ofsocial network analysis. The content of link prediction research includes both the networksclosely related to people’s daily lives and the networks had important scientific studysignificance. Link prediction research has practical and theoretical research importance, andwith the introduction of link concept, link prediction is now becoming hotspots of this area.Meanwhile, link prediction has been widely used in a variety of areas.At the present time, network topology is the main sources of information to solve linkprediction problems based on the similarity. But existing prediction algorithms can not fullyuse the information inherent in the topology, leading to low prediction accuracy. Algorithmcommon neighbors only consider the number of common neighbor nodes, ignoring the linkbetween these nodes, so the algorithm common neighbors can not distinguish between twonodes with the same number of common neighbors, the similarity of the link predictionresults were not accurate. To solve this problem, a new parameter named individual effectcoefficient is proposed. Individual effect coefficient is designed to evaluate the interactionbetween common neighbors, and it reflects the common neighbors set intensity. In addition,based on individual effect coefficient, this paper proposes a new link prediction algorithmbased on the similarity----common neighbors based on individual effect coefficient. Thealgorithm of common neighbors based on individual effect coefficient is a combination ofindividual effect coefficient and common neighbors. The proposed algorithm exploits theinformation of links between common neighbor nodes in the network topology featuresinformation, making the link prediction accuracy increased.Finally, this paper conducts the experiment feasibility and validation for the algorithm ofcommon neighbors based on individual effect coefficient. Making performance comparisonwith several algorithms, to compare the running time and forecast accuracy. Experimentalresults show that the algorithm of common neighbors based on individual effect coefficientimproves prediction accuracy, and the running time is still in the same order of magnitude ofclassical algorithms. The proposed algorithm makes a good balance between the predictionaccuracy and computational complexity.