Which Node Pair and What Status? Asking Expert for Better Network Embedding

Longcan Wu, Daling Wang, Shi Feng, Kaisong Song, Yifei Zhang, Ge Yu

Abstract: In network data, the connection between a small number of node pair are observed, but for most remaining situations, the link status (i.e., connected or disconnected) of node pair can not be observed. If we can get more useful information hidden in node pairs with unknown link status, it will help improve the performance of network embedding. Therefore, how to model the network with unknown link status actively and effectively remains an area for exploration. In this paper, we formulate a new network embedding problem, which is how to select valuable node pair (which node pair) to ask expert about their link status (what status) information for improving network embedding. To tackle this problem, we propose a novel active learning method called ALNE, which includes a proposed network embedding model AGCN, three active node pair selection strategies and an information evaluation module. In this way, we can obtain the real valuable link statuses information between node pairs and generate better node embeddings. Extensive experiments are conducted to show the effectiveness of ALNE.