A central task in heterogeneous information networks (HIN) is how to characterise an entity, which underlies a wide range of applications such as similarity search, entity profiling and linkage. Most existing work focus on using the main features common to all. While this approach makes sense in settings where commonality is of primary interest, there are many scenarios as important where uncommon and discriminative features are more useful. To address the problem, a novel model COHIN (Characterize Objects in Heterogeneous Information Networks) is proposed, where each object is characterized as a set of feature paths that contain both main and discriminative features. In addition, we develop an effective pruning strategy to achieve greater query performance. Extensive experiments on real datasets demonstrate that our proposed model can achieve high performance.
展开▼
