Word embedding-based relation modeling in a heterogeneous information network

Heterogeneous information networks are widely used in big data applications. These networks consist of multi-type information objects and relations. The appearance of the network can be changed depending on what perspective is used for modeling. Modeling relations between information objects has attracted recent attention. Although many related works have been proposed, they have limitations: they are hard to apply to unstructured data and they require continuous learning; and the results are often sparse. In this paper, we propose a new method based on a word-embedding technique that deduces various relations between information objects. We create viewpoint data that reflects any perspective on information objects and word embedding carried out by using these data. Using the proposed method, the system quantifies the relations between the information objects in heterogeneous information networks. The experiments use real world data to demonstrate the effectiveness of our methodology.