Date of Award
Heterogeneous networks, Network embeddings, PageRank, Random walk
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.
Academic networks are derived from scholarly data. They are heterogeneous in the sense that different types of nodes are involved, such as papers and authors. This dissertation studies such heterogeneous networks for measuring the academic influence and learning vector representations of authors. Academic influence has been traditionally measured by the citation count and metrics derived from it. PageRank based algorithms have been used to give higher weight to citations from more influential papers. A better metric is to add authors into the citation network so that the importance of authors and papers are evaluated recursively within the same framework. Based on such heterogeneous academic networks, we propose a new algorithm for ranking authors. Tested on two large networks, we find that our method outperforms the other 10 methods in terms of the number of award winners among top-ranked authors. We further improve the method by finding and dealing with the long reference issue. Moreover, we find the mutual citation in paper networks and the self citation issue in author networks. Our new method can reduce the impact of the above three issues and identify more rising stars. To learn efficient author representations from heterogeneous academic networks, we propose a new embedding method called Stratified Embedding for Heterogeneous Networks (SEHN) based on Skip-Gram Negative Sampling (SGNS). We conduct Random Walks to generate the traces that represent the structure of the network, then separate the traces into different layers so that each layer contains the nodes of one type only. Such stratification improves embeddings that are derived from the mixed traces by a large margin. SEHN improves the state-of-the-art Metapath2vec by up to 24% at a certain point. The efficacy of stratification is also demonstrated on two classic network embedding algorithms DeepWalk and Node2vec. The results are validated in two heterogeneous networks. We also demonstrate that SEHN outperforms the embedding of homogeneous author networks that are induced from their corresponding heterogeneous networks.
Zhao, Fen, "Study of Heterogeneous Academic Networks" (2021). Electronic Theses and Dissertations. 8513.