PROJECT TITLE :

CAMU Cycle-Consistent Adversarial Mapping Model for User Alignment across Social Networks

ABSTRACT:

The user alignment problem is a fundamental issue that arises in a variety of social network analyses and applications. This problem involves establishing a correspondence between users on different networks. Since symbolic representations of users are prone to sparsity and noise when computing their cross-network similarities, the most cutting-edge methods embed users into the low-dimensional representation space, where their features are preserved, and establish user correspondence based on the similarities of their low-dimensional embeddings. This is done because symbolic representations of users are problematic in computing their cross-network similarities. A great number of embedding-based methods begin with the goal of learning a mapping function in order to align the latent spaces of two networks before computing similarities. However, the majority of them learn the mapping function largely based on the limited labeled aligned user pairs, and they ignore the distribution discrepancy of user representations from different networks. This can lead to the overfitting problem, which can affect the performance of the model. We propose a cycle-consistent adversarial mapping model to establish user correspondence across social networks as a solution to the problems that have been outlined above. The model is taught mapping functions that are applicable across the latent representation spaces. The representation distribution discrepancy is addressed by means of adversarial training between the mapping functions and the discriminators, in addition to cycle-consistency training. In addition, the training process for the proposed model makes use of both labeled and unlabeled users, which has the potential to alleviate the overfitting problem and reduce the number of labeled users that are required. The effectiveness of the proposed model on user alignment on real social networks was demonstrated by the results of extensive experiments that were conducted.