The Identification of Influential Users Based on Semi-Supervised Contrastive Learning

Identifying influential users in social networks is of great significance in areas such as public opinion monitoring and commercial promotion. Existing identification methods based on Graph Neural Networks (GNNs) often lead to yield inaccurate features of influential users due to neighborhood aggregation, and require a large substantial amount of labeled data for training, making them difficult and challenging to apply in practice. To address this issue, we propose a semi-supervised contrastive learning method for identifying influential users. First, the proposed method constructs positive and negative samples for contrastive learning based on multiple node centrality metrics related to influence; then, contrastive learning is employed to guide the encoder to generate various influence-related features for users; finally, with only a small amount of labeled data, an attention-based user classifier is trained to accurately identify influential users. Experiments conducted on three public social network datasets demonstrate that the proposed method, using only 20% of the labeled data as the training set, achieves F1 values that are 5.9%, 5.8%, and 8.7% higher than those unsupervised EVC method, and it matches the performance of GNN-based methods such as DeepInf, InfGCN and OlapGN, which require 80% of labeled data as the training set.