INAMA: An Interactive Attentional Model for Node Alignment

Abstract

Because of wide studies of Social Network Analysis (SNA), identifying users from heterogeneous platforms, also known as node alignment, has gradually become a research hotspot. In this paper, we propose an INteractive Attentional Model for Node Alignment, namely INAMA. To tackle the issue, the model leverages both topology structures and node attributes. First, we define the matched neighbors instead of the original topology structures, which consist of neighbors from the aligned pairs. By doing so, our model can efficiently leverage topology information. Then, an interactive attentional model is built to model node message passing processes. Specifically, intra and inter attentional mechanisms are introduced to determine the neighbor influences from local and across networks, respectively. Finally, we evaluate our model on six real-world datasets and the experimental results demonstrate the effectiveness of our model.