Research On Social Network Alignment Based On User Naming Conventions

Social network alignment is an important part of online social network research,and it has vital significance in the practical application of information retrieval,crossplatform recommendation system,network security and other fields.Due to the lack of social network user attributes and incomplete network structure extraction,usernamebased social network alignment algorithm has received more attention.This only focus on the username of social network and study the naming convention of users to solve the problem of account alignment.The main research work and innovations are as follows:(1)In view of the problem that the existing simple classification algorithm in the study of account alignment based on username is easily affected by the imbalance of positive and negative instance training.This paper proposes a social network alignment algorithm based on user naming convention mapping learning.This algorithm realizes mapping between username vectors of two social networks through BP neural network,transforms the classification problem into mapping problem between vectors.In this model,the principle of back propagation in BP neural network is used to adjust the weight repeatedly to learn and train the required mapping function more accurately.Finally,the distance difference between the two vectors is calculated to confirm whether the two usernames belong to the same person,so as to realize account alignment across social networks.The experimental results on multiple social network datasets show that,compared with the baseline methods,the precision@1 of proposed model's account alignment improves by 4% compared with the algorithm UISN-UD in the baseline methods,and the experiment with a smaller training set ratio and iteration times has higher accuracy and faster convergence than the baseline methods.(2)To solve the problem that a large amount of labeled data is needed for supervised or semi-supervised training in the existing account alignment research based on username.This paper proposes a social network alignment algorithm based on user naming convention adversarial learning.Since there is some difficulty in data acquisition for social network accounts with known corresponding relationships,the algorithm introduces the idea of weak supervision training.The algorithm only uses a small amount of labeled training data to project one social network vector to another social network by generating the game mechanism of generator and discriminator in generative adversarial networks.The distribution difference between the two vectors is calculated through the optimal transmission distance,so as to judge whether the two usernames belong to the same person and realize account alignment across social networks.Experiments on four real social network datasets show that precision@1 is more than 25% better than the best baseline method when multiple social network datasets have a training set ratio of 5%,and 16% better than the best baseline method when the training set ratio is 10%.This algorithm can maintain performance with less labeled data.