Research And Application Of Similarity Measure Based On Heterogenous Information Network Embedding

Due to the flexibility of data heterogeneity modeling,heterogeneous information networks(HIN)are used to characterize complex and heterogeneous auxiliary data in retrieval systems,and methods for accomplishing retrieval tasks through similarity metrics have received extensive attention.At present,the HIN-based similarity measurement method generally measures the similarity of nodes through the meta-path.However,for the similarity measure of the same type of nodes,the similarity method based on HIN meta-path has some problems: they ignore the impact of the similarity of other types of nodes in the meta-path to the similarity of the target-type nodes.Aiming at this situation,this paper proposes a similarity measure algorithm based on heterogeneous information network embedding,called HNESim.In order to better embed heterogeneous information network,HNESim uses an improved Deepwalk algorithm: Design a random walk strategy based on meta-path,and generate meaningful network embedded node sequence through node type filtering.The learned node embedding is first transformed by the fusion function,and then the similarity between the two nodes is measured by calculating the cosine values of the two nodes embedded in the fusion.We experimented with authoritative research project management data and compared it to the classic similarity measure method,PathSim.The experimental results show that the average accuracy(MAP)of HNESim is higher than that of PathSim,which reduces the number of unrelated results and proves the effectiveness of HNESim.In addition,HNESim is applied to the collaborator retrieval system.We deeply analyzed the needs of researchers seeking collaborators,and defined the functional module for these requirements,and completed the architecture design,and achieved accurate retrieval of researchers in the collaborator retrieval system.The application of HNESim algorithm in the collaborator retrieval system has solved the problem that researchers are difficult and inefficient to seek collaborators,and has good research and application value.