A detailed technical account, this paper presents a new method of handling heterogeneous information network analysis.
The purpose of the reported research is to reduce uncertain information in the interpretation of heterogeneous information networks, which occurs due to missing values and noise. The authors develop a fusion technique called fusing reconstruction (FRec). This technique converts uncertain characteristics into a homogeneous feature space by fusing heterogeneous features and fusing heterogeneous constraints. FRec is designed with two phases, each one handling one of the heterogeneities. The first phase--invertible fusing transformation (IFT)--is responsible for the ability “to learn latent homogeneous feature representations for heterogeneous nodes” and for transforming the space back and forth in a bidirectional manner. The second phase--heterogeneous constraints fusion based tensor reconstruction model (HCF-TRM)--processes “uncertain snapshots of a dynamic [heterogeneous information network, HIN] and recovers the missing values by fusing the spatial [and temporal] smoothness constraints into the tensor reconstruction.”
The paper presents a through explanation of the proposed approach, by first explaining in great detail the specifics of the heterogeneous information networks, and by motivating the need for creating the presented approach. Further, the outline of the method is given with preliminaries and definitions in a strictly formal manner. The algorithms are presented as pseudocode and thus are easy to understand and implement.
Aside from presenting the new method, the authors describe a set of experiments carried out with the implemented method, and report results showing the performance and the advantages of the new method, based on several evaluation criteria relevant for estimating efficiency and comparing with other methods.
The paper unconventionally discusses related work, in its final part, by grouping the related subject areas into three groups--graph embedding, low rank approximation tensor, and heterogeneous network analysis--and refers to a considerable number of references.
Written with great precision and attention to detail, this work is very good reading for engineers, scholars, and students interested in the theory and practice of information network analysis.