Mining latent academic social relationships by network fusion of multi-type data

Abstract

The relationship extraction and network fusion are hotspots of current research in social network mining. Since the types of data are manifold, researchers can utilize multi-type data to construct multiple networks. In academic social network mining, present researches are mostly based on the single-type data, e.g., the co-authorship network constructed by academic co-authorship records. However, the relationships portrayed by single-type data are not sufficient to characterize the complex relationships of the real world. To solve this problem, we are the first to the best of our knowledge to use acknowledgment text to construct a semantic information-based academic social network. First, we extract named entities from multi-type data and implement network optimization and alignment. Second, a semi-supervised fusion framework for multiple networks (SFMN), using the gradient boosting decision tree algorithm to fuse the information of multiple networks into a single network, is proposed in this paper. Third, we implement the parallel version of SFMN on Spark to improve the performance of large-scale social network analysis. Experiments show the superiority of our framework over several state-of-the-art methods and prove our method can effectively integrate network information.