Psychology-Informed Network Alignment

Nowadays, social networks are booming at a very fast rate, and have spawned many cross-network tasks, such as user portraits, user interest mining and possible friend recommendations. As an important prerequisite for cross-network tasks, the accuracy of network alignment is particularly important. Social network alignment refers to the combination of users who are the same person in different social networks. Traditional network alignment methods use user behavior information, such as network structure, attributes and personally generated content, but they ignore behavior is not as stable as we think. All behaviors are guided by personality, and according to psychological research, personality has cross-time continuity and cross-situational consistency. Due to the stability of personality information, we intend to use personality information to guide network alignment. In this work, we propose a novel multitask framework for enhancing network alignment with a personality recognition task. The doc2vec method is used to extract the features of each user generated content. After the network structure and text features are fused by GNN, the network alignment task and personality prediction task are carried out respectively. Towards evaluation, we create a user identity linkage dataset with enough user generated content collected from Twitter and Instagram. Extensive experiments validate the superiority of the proposed methods.