Research On Link Prediction Based On Network Local Structure

The study of distribution functions of the within-class distance and the inter class distance is of great practical significance.In this paper,we mainly use parameter statistics to determine the distribution function of the within-class distance and the inter class distance.Based on distribution functions of the within-class distance and the inter class distance,we propose a text clustering evaluation method based on parameter estimation of distances within clusters and determine the critical value of ratio type index based on the within-class distance as well as the inter class distance.Firstly,the text clustering evaluation method based on parameter estimation of distances within clusters is studied in this paper.We not only find that this method is feasible while the number of clusters is too small or at the same time as the true class number,but also it can weaken the influence of initial class center selection on K-means algorithm,improving the accuracy of clustering results.Secondly,the critical value of the ratio type index based on the within-class distance as well as the inter class distance is also studied in this paper.According to the experiment,we determined the location of the real value of cluster index in the upper and lower bounds,which canbe used as a basis for judging the ratio type index based on the within-class distance as well as the inter class distance.