Research On Relevance Search Technology Over Signed Heterogeneous Information Networks

In the real world, most of data objects are interrelated or interactive, forming the network which is numerous, interconnected, complex. We explain this interconnection network information network generally, for example, social networking, web, papers cooperation networks, biological networks, etc. Ubiquitous information network is a key part of modern information infrastructure. Information network, or a particular type of it, such as social networking, has become a hot topic among scholars in computer science, social sciences, physics, biology and other fields.Heterogeneous information network is a network where nodes and relations are of different types.In real world, heterogeneous information network has been-widely used, and in many heterogeneous information networks, links have positive and negative polarity to express people’s positive or negative opinions and ideas. For example, IMDB network contains user, movies, actors and other different types of nodes, and the relationship of different types such as the appraisal relationship between user and movie, appeared relationship between film and actor. Users can express their preferences through rating the movies. The heterogeneous information network which add user’s preference model, is called signed heterogeneous information network. The network of links can be positive (expression "like" or "Trust"), or negative (the expression "do not like", "no confidence").Relevance search is an important task of heterogeneous information networks, used to measure the correlation between the different types of nodes in the network, which can support applications such as personalized recommendations. For example, in IMDB networks, people are interested in searching for the most relevant movies with a user, in Epinions network, people prefer searching for the most relevant goods for a consumer. The semantic of different path is not the same. Thus, in previous studies, usually calculate relevance between nodes based on the multi-semantic meta-path, where a meta-path is a path consisting of a sequence of relations defined between different object types.Conventional research on relevance search focuses only on unsigned network without considering polarity of network. However, it is a challenge that the semantic of meta-path with negative sides is defined on signed heterogeneous information network. Especially, while such negative edge is more than one, the semantic of path will be fuzzier. Due to the signed meta-path has positive or negative polarity, we cannot directly use the meta-path-based relevance measure to compute the relevance between objects from different types in the signed heterogeneous information network. Therefore, it is a challenging job to solve such problem.In this paper, a relevance search measure called SignSim is proposed, which can measure the relatedness of objects in the signed heterogeneous information network based on signed meta-path factorization. SignSim firstly defines the atomic meta-paths and gives the computing paradigm of the similarity between objects in 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 from different types based on multi-length signed meta-paths. The main contributions of this paper are as follows:(1) Proposed and described the relevance measure between objects from different types in the signed heterogeneous information network. (2) A relevance search measure called SignSim is proposed, which can measure the relatedness of objects in the signed heterogeneous information network based on signed meta-path factorization. Semantics of the positive side or negative side on the meta-path can be effectively captured by SignSim. (3) Experimental results on two real-world datasets (IMDB, Epinions) verify the effectiveness of our proposed approach by comparison of the existing correlation search method on unsigned network.