Signed Heterogeneous Network Embedding in Social Media

In today's social networks users can express positive or negative attitudes towards others, which forms sentiment links in addition to social links. This link heterogeneity builds a new network topology called signed heterogeneous networks. Meanwhile, network embeddings have emerged as convenient methods for conducting machine learning over networked data. Consequently, extending embedding methods for signed heterogeneous networks allows interesting analytical scenarios such as predicting the sign of sentiment links. Previous works mainly focus on heterogeneous network embedding discarding the sign information in the network. In this paper, we propose SiHet, a fast and scalable embedding method suited for Signed Heterogeneous networks. Our method extends classical network embedding algorithms through a new loss function and a sampling strategy which allows to integrate sign and social network information. Empirical evaluations on two real-world datasets demonstrate effectiveness and efficiency of SiHet compared to the state-of-the-art methods.