Research On Link Prediction Based On Local Features

In the 21 st century,the rapid development of computer technology brings more and more complex networks to our daily life.Complex networks have become an important research field.As a key component of complex network studies,link prediction is thriving vigorously.It can be used to predict the possibility of future connections between nodes in the network that have not yet formed a connection relationship,which is of great significance in solving various problems occurring in real life.For example,we can apply link prediction to product recommendation or to guide protein interaction experiments.We can use the local characteristics(local information and local path)of the nodes in the network to perform link prediction.This thesis makes more efforts to use the local characteristic information of the nodes.The research content is mainly shown as follows:(1)Combining local characteristic information with the deep learning algorithm— Convolutional Neural Network Model,we constructed the algorithm flow — Link Prediction based on Convolutional Neural Network(LPCNN).The core of the LPCNN algorithm flow is the construction method of characteristic matrix.After the algorithm extracts the similarity sequence,it constructs the feature matrix through the feature matrix construction algorithm.Then the feature matrix and the defined labels are sent to the Le Net-LP convolutional neural network model modified for the link prediction task for training and learning.Finally,we evaluate the link prediction performance of the model.The algorithm has achieved good performance on some publicly complex network data sets.(2)We propose ELP similarity index.This index was used to improve the calculation of the link prediction similarity index based on the local path.During the empirical research,we used the selenium crawler framework to collect the actor cooperation data of the TOP25 annual box office champs from 2016 to 2020.Based on the collected data,we constructed the Actor?TOP25 network dataset.Compared with similarity indices such as CN,AA,RA,and LP,ELP index shows a higher accuracy of link prediction classification when testing on the Actor?TOP25 dataset.(3)We applied the local path information to the binary classification model in machine learning,and proposed Link Prediction based on Support Vector Machine(LPSVM)— a link prediction algorithm on the basis of the support vector machine model.One of the innovations in this algorithm is that it constructs new features,which makes full use of the high-order paths information between sample pairs.The algorithm presents a superior performance of the link prediction classification on four empirical datasets.