Relevance search is a primitive operation in heterogeneous information networks, where the task is to measure the relatedness of objects with different types. Due to the semantics implied by network links, conventional research on relevance search is often based on meta-path in heterogeneous information networks. However, existing approaches mainly focus on studying non-signed information networks, without considering the polarity of the links in the network. In reality, there are many signed heterogeneous networks that the links can be either positive (such as trust, preference, friendship, etc.) or negative (such as distrust, dislike, opposition, etc.). It is challenging to utilize the semantic information of the two kinds of links in meta-paths and integrate them in a unified way to measure relevance.
In this paper, a relevance search measure called SignSim is proposed, which can measure the relatedness of objects in signed heterogeneous information networks based on signed meta-path factorization. SignSim firstly defines the atomic meta-paths and gives the computing paradigm of similarity between objects with the same type based on atomic meta-paths, with collaborative filtering using positive and negative user preferences. Then, on basis of the combination of different atomic meta-paths, SignSim can measure the relatedness between objects with different types based on multi-length signed meta-paths. Experimental results on real-world dataset verify the effectiveness of our proposed approach.