IIIT Hyderabad Publications |
|||||||||
|
Towards Mining Intellectual Influence Associations and Developing Efficient Author Representations in Bibliographic NetworkAuthor: Tejas Shah Date: 2020-06-09 Report no: IIIT/TH/2020/45 Advisor:Vikram Pudi AbstractWithin the academic scholarly environment, the contribution of an author in the form of publications holds an intrinsic value responsible for the effective dissemination of knowledge for the scientific community. In such a social system of science, citation practices characterize social functions like the conferral of recognition upon the work of others as well as the acknowledgement of one’s intellectual debt. Citations and references operate in a collective cognitive and moral framework designed to provide historical lineage of knowledge and repayment of the intellectual debts through their acknowledgement. However, the structure of intellectual influence is misrepresented when only the immediate citations and their cardinality are taken into consideration. Citation analysis has the potential in providing valuable insights about the social system of science, including the hierarchical and fractal nature of scientific development spanning across a wide range of topics. Thus, in order to better understand the associative dissemination of influence and approximately construe the anatomy of this structure, complex interactions in the convoluted network of authors and papers need to be probed. Our study aims at understanding these influence associations eventuated between authors based on citation data in the heterogeneous bibliographic network. Besides learning the existence of such associations, our work focuses on quantifying the extent of the pairwise impact measure between authors. For the bibliographic dataset of authors and publications, we define proxy scores that attempt to determine the associative bilateral influence of the cited author over the citing author using a diffusion-based influence aggregation approach. Since the ranks of authors range across a broad spectrum, we present a detailed qualitative analysis and suggest suitable extensions for variants in academic networks. For assessing predictive capacity and competency of the devised scores, we develop an objective approach of evaluating the influence model for the bibliographic dataset. We perform an empirical analysis and validation using author representations derived by using the proposed bilateral influence scores and the baseline weights. These representations are generated by profiling author interactions within the bibliographic network using representation learning. Subsequently, we discuss the experimental results subject to a classification task along with a comparative study against observations with those obtained with the baseline. The latent vector representations of authors developed harnesses the structural connectivity and capture the network topology of the academic network. While the efficacy of these representations in the context of the problem statement is discussed, they form a generic multivalent result. It constitutes the underlying mechanism of the proposed influence model and suggests utility for a diverse set of statistical modeling tasks within the bibliographic domain. The results summarized in this thesis significantly deepen our understanding of the underlying influence mechanisms governing the system of science. Finally, our work in this thesis lies in the broader research initiative of citation analysis and aims to contribute to the literature on the identifying and quantifying influence associations in the scientific network. Full thesis: pdf Centre for Data Engineering |
||||||||
Copyright © 2009 - IIIT Hyderabad. All Rights Reserved. |