Heterogeneous Information Network Hashing For Fast Nearest Neighbor Search

Heterogeneous information networks (HINs) are widely used to model real-world information systems due to their strong capability of capturing complex and diverse relations between multiple entities in real situations. For most of the analytical tasks in HINs (e.g., link prediction and node recommendation), network embedding techniques are prevalently used to project the nodes into real-valued feature vectors, based on which we can calculate the proximity between node pairs with nearest neighbor search (NNS) algorithms. However, the extensive usage of real-valued vector representation in existing network embedding methods imposes overwhelming computational and storage challenges, especially when the scale of the network is large. To tackle this issue, in this paper, we conduct an initial investigation of learning binary hash codes for nodes in HINs to obtain the remarkable acceleration of the NNS algorithms. Specifically, we propose a novel heterogeneous information network hashing algorithm based on collective matrix factorization. Through fully characterizing various types of relations among nodes and designing a principled optimization procedure, we successfully project the nodes in HIN into a unified Hamming space, with which the computational and storage burden of NNS can be significantly alleviated. The experimental results demonstrate that the proposed algorithm can indeed lead to faster NNS and requires lower memory usage than several state-of-the-art network embedding methods while showing comparable performance in typical learning tasks on HINs, including link prediction and cross-type node similarity search.